Instead of picking a target UUID and evaluating new UUIDs against it, a better experiment would be finding duplicates in all the UUIDs you have generated.
It would require a lot more memory, because you have to remember every generated UUID. And how would you do the partial match? You are not going to observe any collisions.