I would guess that this aspect of the spec was not well tested. It is possible to write a test that causes 100 or 1000 collisions and plots the random distribution of the backoffs, but that is pretty complicated versus just checking that 1 collision had a backoff that was within range.