Yes. In reality, we are talking about populations. It is still possible that the type of confounded you describe applies at the population level as well, which

Yes. In reality, we are talking about populations. It is still possible that the type of confounded you describe applies at the population level as well, which is why I qualified my statement with "might expect". This is an empirical question that requires empirical evidence to answer. Further, universities have this empirical evidence (although I am not aware of what it actually says). I do recall hearing that (at some point in history) the army looked into this and found that black soldiers did better then their white counterparts with an equal score entering. Further, this is presisly the result I would expect to see in the face of generic anti-black bias. For instance, if there was some force that depressed black SAT scores by 10 points and had no other effect, then a black student would be expected to perform as well as a white student who scored 10 points better on the SAT.