The argument that it doesn't matter how racially biased a test is because it's good in other ways isn't something I am inclined to accept.

I assume this is hyperbole, since obviously a truly perfect test could draw from any subset of the population, as long as it was large enough to contain near-perfect individuals.

With that said, I agree, we should attempt to avoid any bias in such a test, including that of race (I would not, however, single this possibility out.) That is what I meant by

That said, you would definitely have to be careful to ensure the test was as good as possible.

However, beyond a certain level of conscientiousness, demanding perfectly unbiased tests becomes counterproductive; especially when one focuses on one possible bias to the exclusion of others. In truth, even age is a racially biased criterion.