Certainly not enough different to actually prove anything.

You can check in even more depth by seperately analysing them for each of his big winning sessions - if it's ALWAYS below average then that starts to look suspicious.  You should ignore losing sessions when doing that - just the half dozen (or however many it is) sessiosn where he won big.

There's no selection bias in that as we're choosing samples based only on the results of the big bets.  If it's genuinely random then you'd expect some sessions to have small bets winning more often than 49.5% and others losing more often - that overall it ends up slightly below isn't so important.

Think of the overall result (quoted) as meaning he flipped more heads than tails in a series of coin tosses (which doesn't mean a lot).  Now we're checking to see if they were ALL heads (which means a lot more).

Similar analysis can't be applied to the big bets - as his strategy (win decently - or lose big) adds selection bias.  But his ending points weren't chosen based on the performance of the small bets so they can be analysed in a manner that can't be done for the big ones.

Reason I say to only look at big winning sessions is simple - if he IS cheating then he'd throw some losing sessions in intentionally where the small bets (and big ones) would likely be random.  Incidentally, if he worked out a pattern then I don't personally regard it as cheating.  Cheating would be if he somehow got hold of server seeds.