Assuming what the spirit of what la Griffe says is true (and ignoring the annoying format)
I think he dramatically overstates the generality of his case.

In particular he may have found the maximum productivity, but if the maximum is very flat then the change in productivity vs the ability of the worker may not be worth the price of the more productive workers. His theory would work best in a school situation where everyone is competing individually against everyone else. In a business situation where everone works together to produce widgets, because of my assumed flatness of the maximum I would postulate that it would be possible to come with different distributions of worker ability for each job to get only slightly less productivity as the la griffe method with less brain power and hence lest cost. That is each job discription can handle a certain % of dumb people who can imitate the smarter people doing the job type. These effects arise from the fact that total sum can be greater then the sum of the parts.

I think this is one of the points made in the bell curve where they talked about how having all the smart people go to college has affected the efficency of non college educated ocupations in america. I think the bell curve authors would find it ideal that some portion of the college eligible crowd doesn;t go to college and take college type jobs.

I think he applies the idea of Bayesian updating to a case where it obviously applies. We have good population data, and we have updates based on test scores. Though it seems most unfair, purely 'future score maximizing' logic would discriminate against underperforming minority groups. Harsh. But this just highlights the degree of intereference caused by going the other way, and giving bonus points for having particular grandparents.

no name-
I think that for the factory/widget example, La Griffe makes the reasonable assumption that all the workers would be offered the same wage, so that the difference in costs would not be an issue. Also, from a statistical perspective, the difference in productivity is marked, so I question the flatness assumption.
If we take La Griffe's arguments to be true (I have some objections, see below -) then this shows yet another point at which utilitarian philosophy and utilitarian ideals diverge. It should also be pointed out that there would be severe negative externalities associated with allowing implementation of such policies, which La Griffe does not delve into.
But on to the question of whether this argument is valid. It seems to me that La Griffe assumes that the reason that SAT scores and GPA are not perfect predictors of school success is due to random variation in the scores achieved (the effect of luck). So, in a given applicant pool the high-scoring 'low baseline' minorities are assumed to be statistical outliers who typically have a large 'luck' component in their high scores, and the regression of their performance to the mean for their groups is very likely.
I dispute this assumption, based on my own personal experience. I think that the SAT is a very reliable indicator of intelligence and scholastic aptitude, and that a high score is unlikely to be the result of luck. The reason it is not a perfect predictor of college success has to do with its failure to measure attributes such as perseverance and good work habits (I speak as someone who scored 1520 on the SAT at the age of 15 and subsequently dropped out of college). So an alternate model would hold that the SAT does a perfect job of evaluating intellectual fitness for college, and that the other attributes required have not been shown to vary between groups (at least in La Griffe's analysis), so that using it without the differential cutoff makes perfect sense. In this case I think the assumption of random noise in the test scores has led him astray.

substitue 'liberal ideals' for 'utilitarian ideals' in my above post. teach me not to proofread..

Bbartlog, I think your theory is testable in a number of ways. For instance, what would be interesting about the statistical outlier question is whether for different ethnic groups there is a different degree of repeatability in higher SAT score results. e.g. if you took a bunch of people who tested with 1400 total SAT scores and retested them would there be a lower average score for some ethnic groups than other ethnic groups in the second round? If luck is operating in the way that La Griffe assumes then you'd expect a drop-off in test scores in some ethnic groups.

Its likely that ETS or someone else has done the sort of study I'm describing. Does anyone know if this is the case?

There's some blogger who either works for ETS or knows the territory real well. I can't recall her name or her blog name. Anyone know if she's still active? She might know of relevant studies.

i think you mean kim swygert of number 2 pencil. she's a respectable lady-i don't know if she'd touch this....

Simply measuring the degree of random variation in people who take the test multiple times would also provide some useful data. But since some people take test prep courses (or just learn more by various means) between tests you'd have to find a way to account for improvements due to increased knowledge, and separate that effect out before measuring the true random variation.