How much selection pressure can you get if you're only selecting at the level of the gamete, and not at the level of the person or the gene? For example, given a man and a woman, suppose that you looked at 1 million of the man's sperm and picked the best (according to whatever criteria you're using), and you picked the best out of 100 of the woman's eggs, and you mated them. If that was how everyone reproduced, would that provide enough selection pressure to satisfy whatever goals you have for the gene pool?

You can get a lot of pressure. Steve Hsu in one of his slides on BGI (which I can't be arsed to look up) gives a quick estimate that you can get maybe half a standard deviation per generation at ordinary embryo fertilization numbers.

This is plausible if you think about it in a very rough sort of way.

We know IQ is something like 50% genetic, right? If I make 5 embryos for a couple, that's like them having 5 kids; if you saw a normal family of 5 kids, how much smarter than the average of the 5 will the smartest of the 5 be? At a guess, 20 points sounds too much to often happen, but 5 points sounds way too little, so maybe 15 points; we said half the variance was genetic, so that suggests the underlying genes can claim credit for 7-8 of the excess points. If the couple has only 1 kid and the embryo with the best genetic scores is picked, then the genetic base will go up by ~7 points. And as it happens, one standard deviation is often 15 points so 7/15=0.5 standard deviations. Rinse and repeat for the next generation.

EDIT: http://duende.uoregon.edu/~hsu/talks/ggenomics.pdf Gives an example of selection on 100 alleles etc for an estimate of ~0.2SD per generation in that scenario; the more alleles you select on, presumably the higher you get per cycle.