In any case, with such a small sample, it seems quite a high deviation from what the mean should be, I agree.
Is it possible that not all winners came to the Bitcointalk forum and reported their winnings? 5 or 10 big wins in this statistic can significantly change the final assessment, and that is quite a possible scenario.
That is another point. It is a sample of those who reported the winnings. Not a record of casino rolls.
But that's the thing. Even if we had an exact record of all casino rolls, statistical improbabilities are still possible. Randomness is a key part of gambling, after all.
The point examplens raises about unreported wins is very valid. A small sample size, especially with missing data, can definitely skew the overall picture. But, while a complete record of every roll would be fantastic for analysis, it wouldn't eliminate the possibility of outliers entirely. Actually, that's why martingales and similar strategies don't work well in gambling. Even with a perfect 50/50 chance, there's still a very real possibility of hitting a losing streak of twenty, or even more, rolls.