It's a histogram. It's counting the number of buckets (y axis) for each number of collisions (x axis). He's showing that the distribution is the same for both methods; if one had a clear advantage, it should have a relatively more uniform distribution.

The problem is that the input distribution is uniform, so you'd expect a pretty decent distribution of collisions regardless of what your modulus was. A prime modulus only helps for non-uniformly distributed inputs.