[jimmyfingers]

one problem with your model is it just looks at the final result. so if the person goes below 0 and then comes back it counts that as a positive amount instead of the 0 it would be in reality (since the person would stop playing after hitting 0).

to correct this you'd have to model it in smaller chunks. for example, instead of using the final results for 10000 hands, run 10000 individual hands, summing after each one, and if it ever goes below 0 end the simulation. this would represent one trial of 10000 hands.

with a very high standard deviation, of course there will be some huge wins, but there's a very good chance you went bust before getting those wins.

edit: this will not make a big difference if the probability of busting is ~= 0 (low standard deviation, high starting bankroll, and/or small number of hands).