Discrimination is an interesting thing, look hard enough and you'll find it whether or not it exists.
An alternate explanation for what they are finding in that paper: The creators of the test favor questions whose response correlates well with how the person scores overall. A question that low-scoring people are more likely to get right is a bad question.
What the researchers should have done is looked at racial differences in answering questions given the same overall score. Of course, that won't find discrimination so they're not going to publish it.
Your explanation could also be true and at the same time feeding into the same problem if the people that have done well historically have done so because of test bias.
Your proposed experiment doesn’t test what you think it does. Since we are trying to determine if the test is biased you can’t control using the score on the test that might be biased.
