Essentially, the performance gap disappeared when the students formed study groups ("learning groups" in this article). The students were not selected from the general population, but from UC Berkeley undergraduates. Let's see if anyone can replicate these results on a large scale in average American public high schools.