Differences in SAT test scores are the result of innate differences.

The SAT mostly measures g, the highest-order factor common to all subtests in a large and diverse battery of mental tests.

Scholastic Assessment or g? Psychological Science 15, 337 (2004)

The SAT's correlations with 7 other traditional IQ tests in this study a non-weighted average 0.74. Its correlation with Raven's Progressive Matrices, a nonverbal test of inductive reasoning that is widely accepted as a marker of g, is 0.483 (0.72 after correction for restriction of range in ability in this college sample). Based on these data the g loading of the SAT (that is, the correlation of the SAT with the latent construct measured by all IQ tests) probably exceeds 0.80.

g is highly heritable. At least 70 percent of its variance is attributable to genetic differences in the population.

Sources of human psychological differences: the Minnesota Study of Twins Reared Apart. Science 250, 223-228 (1990)
Large, Consistent Estimates of the Heritability of Cognitive Ability in Two Entire Populations of 11-Year-Old Twins from Scottish Mental Surveys of 1932 and 1947. Behavior Genetics 35, 525-534 (2005)
A Quantitative Genetic Analysis Of Cognitive Abilities During The Second Half Of The Life Span. Psychological Science 3, 346 (1992)

The narrow-sense heritability of g/IQ (that is, the component of variance transmitted by parents to their offspring) is probably lower than 0.70 but is still substantial.

Nature, Nurture, And Cognitive Development From 1 To 16 Years. Psychological Science 8, 442 (1997)
Chapter 4 in Intelligence, Heredity and Environment. eds. Robert J. Sternberg & Elena Grigorenko (Cambridge University Press; January 28, 1997)

Look at Figure 3 of this paper and note the extremely weak IQ correlation between biologically unrelated adoptive siblings together (r = 0.04, 398 pairs):

Genetic and environmental influences on human psychological differences. J. Neurobiol. 54, 4-45 (2002)

Whether the sexes differ in g is a controversial matter that is currently being disputed. It is clear, however, that they do differ in other ability factors measured by standardized tests (e.g., spatial-visualization). Moreover, these other ability factors show incremental validity in the prediction of academic outcomes such as choice of field.

Importance of Assessing Spatial Ability in Intellectually Talented Young Adolescentes: A 20-Year Longitudinal Study. J.Edu.Psychol. 93, 604-614 (2001)

Individual differences along these dimensions are heritable independently of g. The most recent analysis of their heritability known to me is contained in this paper:

Sex differences in mental abilities: g masks the dimensions on which they lie. Intelligence doi:10.1016/j.intell.2006.03.012 (2006)

Note that the shared environment parameter was dropped from the quantitative-genetic model of the determinants of the lower-order ability dimensions without significant deterioration of model fit. This makes it rather unlikely that whatever causes between-sex differences can be accounted for by environmental differences that vary across families, which include just about every variable that culture-only advocates have invoked.

I think the fact that a 1 month test-taking course can bump a score by over 100 points scuttles that. As the ETS itself says "it is a myth that a test will provide a unitary, unequivocal yardstick for ranking on merit." SAT scores also over predict future male performance and underpredict femal performance in university.

Companies like Princeton Review and Kaplan grossly overhype the effectiveness of their coaching. 100 points is about the most that their programs can boost your score. However, some individuals would have attained gains of that magnitude on their own (because of measurement error or some true change in ability level owing perhaps to maturation), others actually do worse, and others see no effective change. The average gain over all coached individuals is something on the order of 30 points. A review of practice effects on the SAT is included here:

Straight Talk About Mental Tests. Arthur R. Jensen (Free Press; 1983)

A more recent study is in accord with past findings:

Effects of Coaching on SAT I: Reasoning Test Scores. J.Edu.Measurement 36, 93-118 (1999)

As for differential prediction for males and females, this is confounded by the fact that males and females tend to major in different areas. Statistically adjusting for this disparity substantially reduces differential prediction.

Students perform lower if their expectations of their own ability are lowered or they are aware of race/gender differences in test outcomes.

You are referring to "stereotype threat." First of all, these effects are small and cannot possibly account for the observed group diparities that they are often invoked to explain away. For example, in the MISTRA study referenced above, the male advantage on tests of spatial and mechanial ability exceed one standard deviation, which is far larger than the effect size of stereotype threat. See the highly critical commentary on this research here:

On Interpreting Stereotype Threat as Accounting for
African American–White Differences on Cognitive Tests. American Psychologist 59, 7-13 (2004)

2. The sciences as a whole are made up of people that fall at the far-right edge of whatever cognitive ability curve you're talking about. (And you have to do better than anecdotes from particular disciplines, like spatial grasp and certain engineering fields.)

I claim that this has been established beyond any reasonable doubt. Please read the article that I linked to in my previous comment for an introduction to the oceanic literature on this topic. For a more recent article:

Tracking Exceptional Human Capital Over Two Decades. Psychol.Sci. 17 194 (2006)

Also, as a sort of experiment, visit the local math department of your nearest top-20 university and try to find out the GRE-Q scores of the graduate students there. I predict that not a single one will have a score one standard error below perfect (about 770). It is even likely that not a single one failed to score a perfect 800.

3. Academic science is purely meritocratic, and having a bit more of a particular science-related ability will help you have a more successful scientific career (as opposed to, say, the ability to manage a lab full of students and postdocs, grant writing skills, work ethic, ability to get your name on papers with minimal input/effort, writing a lot of reviews, being buddies with journal editors, being a convincing and charismatic advocate of your ideas at conferences, etc, etc...)

No doubt all these factors orthogonal to cognitive abilities do contribute to success in any field. Yet it remains the case that no other single predictor accounts for as much variance in workplace performance and job knowledge as IQ.

Handbook of Understanding and Measuring Intelligence. eds. Oliver Wilhelm & Randall W. Engle (Sage Publications, Inc.; 2004) (chapter by Ones)

If all of these variables could be quantified and entered in a multiple regression with some measure of performance as a criterion variable (e.g., number of publications, tenure/not tenure in a logistic regression), I take it you do not disagree that a measure of cognitive ability would undoubtedly have positive regression coefficients. (Or you shouldn't, given the evidence that I present next.)

4. Relatedly, the higher you go up in academic prestige (tenure, promotion) correlates with (let alone is caused by) innate abilites.

In fact, there is such a correlation. The linear correlation between test scores and criterion variables holds throughout the range of ability, even at the highest levels. The threshold hypothesis implicit in your statement receives no support in the literature that I know of. For a discussion of this matter, see pp. 289-290 here:

The g Factor: The Science of Mental Ability (Human Evolution, Behavior, and Intelligence). Arthur R. Jensen (Praeger Publishers (1998)

For a more recent study, see here:

Creativity and Occupational Accomplishments Among Intellectually
Precocious Youths: An Age 13 to Age 33 Longitudinal Study. J.Edu.Psychol. 97, 484–492 (2005)

Two SMPY cohorts, all in the 99th percentile of SAT-M scores as junior high students, were subdivided into quartiles based on SAT-M scores. Even at this extremely high threshold of mathematical ability, the subjects in the upper quartile of the 99th percentile earned more science/math doctorates, earned more doctorates in all fields, enjoyed higher incomes, obtained more patents, and secured tenure at top-50 American universities with greater frequency than their counterparts in the lower quartile.

why do you find it so easy to dismiss discrimination as a primary cause? Doesn't it just make sense to work harder toward meritocracy and then see where we stand?

I do not dismiss discrimination as a cause. My hypothesis is that it is not the only cause. As for working toward meritocracy, Steven Pinker has put it best on
Edge: The Reality Club:

"I share Nora Newcombe's desire to move away from a concern with gender differences in mathematical ability to a focus on individuals and how we can maximize their abilities, at least in the spheres of education and public policy. But she does not play out the radical implications of this move. Other than in the context of evolutionary psychology (which elegantly predicts a number of interesting gender differences), a focus on gender differences arises because people ask why the genders are disparately represented in certain walks of life. Almost invariably, disparities in numbers are interpreted as proof of discrimination and discouragement. This, of course, is a fallacy, since the disparities could arise from differences in average temperaments and talents instead or as well. And it's a fallacy with consequences: if the discrepancies attributed to bias really come from sex differences, then the costly measures designed to counter them (aggressive affirmative action, presumptions of ubiquitous prejudice, re-education programs, diversity bureaucracies, etc.) are misbegotten. If people didn't obsess over disparities in gender representation in the first place, they would not create the need for researchers to determine whether the disparities may be caused in part by gender differences in ability or interests. So if people want to minimize the importance of the science of gender differences, they should speak out against gender bean-counting in university science departments."

Thank you for your comments.