Razib has talked about this – here’s what I think.
The various sweeping alleles that have made Europeans and North Asians have light skin were not favored because they helped you garner extra vitamin D, at least not mostly. This is apparent from the allelic structure. There is a single sweeping variant for SLC24A5: if paleness was the point, many partial loss-of-function alleles would be favored, rather like what we see with G6PD deficiency (a malaria defense). But there is only one: so loss of function is not the point (or at least not the sole point): that particular variant has some other advantage, a big one. The fact that it’s favored up on the Ethiopian plateau is another sign that it isn’t driven by vitamin D. There is only one blue-eye mutation (OCA2), only one main blond mutation (KITLG). The SLC24A5 mutation has by far the largest effect on skin color – if it isn’t being selected for paleness, the sweeping alleles of other coloration genes aren’t either. Probably not even MC1R – it has many common loss-of-function mutations, which implies that loss of function is favored – but probably not because of the paleness. Something else.
If you’re close enough to the equator, you need dark skin, and the genes in these pathways are constrained – not free to change. But if they are free to change, then a change that improves fitness through a pleiotropic effect is now free to spread, since the resulting paleness is bearable. In many cases that would require a specific change, not just loss of function – so we see an allelic structure unlike that of G6PD.
The same is likely the case with the various sweeping skin-color alleles in East Asia (although they are less colorful).
So we need to figure out what useful thing these mutations actually do. Dropping the vitamin D hypothesis raises lots of questions.