Note that the top 25% put 99 or above for Universe. Of those, I would be surprised if there weren't a big chunk that put 100 (indicating 100 - epsilon, of course). This is not weighed in appropriately. Likewise for the bottom 25% for Galaxy.

Basically, "If you hugely truncate the outside edges, the average probabilities wind up too close together" should be entirely unsurprising.