100 occurrences is far to small a sample size to expect regression will be close to the mean every time. So it's not going to be especially unusual to see 30 or 40 when the expected value is 35. However, if you increase the sample size to say 100,000 the likelihood of seeing 30,000 or 35,000 vs the expectation of 35,000 declines significantly.
That said, there's still a hole in the initial assumption regardless who said it since there's no consideration given to skill. Over the long run, better players will win more money and lose less with the same distribution of hand held.
|