Yes. Some of the first statistical analyses of poker I came across when I was learning were based on simulations using bots run by none other than Nate Silver (he used to be a poker pro). His conclusions were in line with what I've been saying, and these simulations were using full and 6-handed tables if I remember correctly. Heads-up has even more variance.

Er, sorry but I don't suppose you've got a link?

Let's say you came out even after 100,000 hands. If I randomly divided those hands into two sets of 50,000 each, would you expect one of those sets to be up by 10% and one of those sets to be down by 10%?

Wish I did but it's buried deep in the archives of the twoplustwo.com forums. You'd probably stand a better chance of finding it by contacting Nate and seeing if he has a copy of the writeup lying around still than using their godawful search interface.

Not sure I understand your other question. There are no percentages in particular that I would 'expect', but if you ran the sim enough times then yes they'd revert to some mean. I've never done such analysis so I don't know what the results would look like.

K - thanks for answering!

Not sure if you're still following this thread, but another major factor is that only a very small percentage of hands actually have much monetary significance. At a full table in limit Hold 'Em, a strong player will fold ~80-85% of his hands preflop, and of those that go to the flop, only ~40% of those go to a showdown, and only a fairly small percentage of those are medium to large pots.

Luck on these bigger pots has a huge impact on overall win rate, but in a set of 100,000 hands, you may only be involved in 2-3000 of them, and you may only expect to win something like 400 or 500 (very rough estimates). Maybe that helps explain why there's more variance than you might expect in a seemingly large sample.