Well here’s a nasty little result. It turns out that the best way to select for optimum performance from a multi-racial applicant pool is to discriminate against the low-performing groups. You need not follow the rather daunting math to follow the reasoning: the greater the difference between a score and the mean, the likelier it becomes that the score was lucky. If you adopt a single cutoff, the members of low-performing groups who are admitted have a greater differential between their scores and the mean (for their groups) than others. Therefore, their results are more likely due to chance. Therefore, you need to set a higher cutoff for the low-performing groups to compensate for this fact. And therefore people who clamor for a single cutoff for all applicants are not advocating a meritocracy, even though they think so.
(Link from Gene Expression, who else?)