By the way, the paper is a lot of words and math to explain that homogeneous groups can fall into local minima, whereas diverse groups can find better minima. I don't necessarily claim it disproves what you say, but it is at least a paper on the subject.