Just as a side note to all, there is a great deal of published papers on randomness in the artificial intelligence community. Believe it or not, if a chess program is capable of generating "better randomness", they get better use out of their hash tables, which helps to reduce the size of the game tree searched, which can help it get to a deeper search depth in the same amount of time!

I experiemented with different random number generation routines, and, sure enough, in the grand scheme of things, some code produces "better randomness" than others.

But please take a look at the huge number of active gammon games shown on the statistics page. With so many games in play, the odds are, evenly unlikely rolls are going to happen.

As one Quantum Physicist said: "If you wait long enough, everything will happen."