It's possible that backprop has more success for narrower bounded problems and is "faster," but maybe because it's greedy. Biological learning might benefit from being less greedy.