Does 'diversity' in ensembles really work? Comparing different seeds vs. bagging + width resulted in a 3.1x disagreement rate

"The more diverse the members of an ensemble, the better." This is a common phrase in the world of machine learning, but ...