I think there are a lot of distributional prior assumptions going into the model. Eg:
Assumption 1: there will be a lower -IQ producing genotype and then there will be a say a zero-mean noise distribution around it in phenotype (here measured by IQ).
Assumption 2:
Immigrants are selected on higher phenotype but not genotype.
Assumption 3: Breeding is done based on favourable phenotype (if at all --> well-fare state, advanced medicine) not genotype.
Then there is selection bias toward the upper end of the phenotype but of course this leaves the genotype's mean unchanged, so their kids will still not have a greater chance of being smart than the rest of the immigrant's home population and, in particular, the expected IQ of the second-generation immigrants will not be greater than the expected IQ of the home population, in spite of the first generation being measurably smarter (cannot see a contradiction under these assumptions).
That aside, I think Stefan's focus on IQ might skew the debate a little. I know he is an IT guy and I am sure for a coder IQ is a good measure of success. But I do not think the same is true for other meaningful professions (eg lawyers, doctors, entrepreneurs (?) etc).
Now, IQ clearly matters if your society needs a lot of Engineers but otherwise not so much I would argue. This is the reason why IQ is such a contrversial quantity and there are multiple competing tests and scales of IQ.
By the way: Gossip goes that MIT dismisses standardized tests like the GRE (which correlate with IQ) because of the fact that a lot of their top scientists obtained low score and due to the iffy evidence of their general predictiveness.