2) if expectation is to lose 25k, to beat it by 40k you 'only' have to win 15k. That means winning 5018 times (5018*400 - 4982*400 = 14400)
Using your numbers, he would be about 8.8% to win 15K or more. That's based on his winning 5018+ out of 10K rolls, with a 49.5% chance to win each individual roll. That's obviously in the realm of possibility.
My understanding is that he has played more than 10K rolls, but the fact he varies his bet sizes adds more variance. As someone already mentioned, some kind of Monte Carlo simulation would be the best way to figure out how probable his win streak is.
We could run his exact same bets through a simulator 10,000 times and see how often his results are met or exceeded. No mathetmatical proofs required