The latter is a truly wicked problem: we can't find out how things operate under novel circumstances until those novel circumstances are realized. (When you have something like physics where the rules are so well-grounded, you can speculate more confidently than you can with something fuzzy like human society and nature.) It's no sin of a theory to not tell us these things for sure, unless it's rhetorically marshalled for that ("women won't become great scientists because on the savannah...") Although it could count as Bayesian evidence.

With things like sex and racial differences you can happily say (if your values swing that way) "okay, maybe it's that way and maybe it's not, but the best way to put egalitarianism to an empirical test is to realize as complete a social equality as possible and see what happens. Purely in the name of science, of course." What makes the problem really wicked is that some counterfactuals would be very undesirable to implement. We just have to hope it's knowledge we'll never get to obtain.