@fig:
Your fallacy relies on the principle in statistics known as the law of large numbers. It states, essentially, that as the number of trials increase, the proportion of occurrence approach the expected proportion. The thing is, the majority of users on the site don't have enough games to make this true. The smaller the probability of the occurrence, the more trials you need. I'll illustrate this by example:
Let's say the typical user has 20 games. (Probably above average for a user on this site, even after you take out all the people that play one game and leave) Let's say that 5% of games have a cheater. (Again, probably a generous assumption. In all likelihood, I'd say only 1-2%) That means that, for the typical user, only one of their games had a cheater. Either they benefited or they were hurt from that cheater, and in all likelihood, the cheater soloed. So lets say that there's a 50% chance they soloed, 25% chance I benefited, and 25% chance I was harmed. (Again, I think these numbers are more than generous. If you disagree, please say so.)
These numbers mean that, relative to other users, in 1.25% of their games, a user gets helped by a cheater, and in 1.25% of their games, they get hurt by a cheater. In 2.5% of their games, everyone gets screwed. 1.25% of 20 games isn't sufficiently large (In fact, it's only 1/4th of a game) to make things balance out. At most, the typical user will have only one game, and they can only either benefit or get hurt in that one game. (Remember, this is ignoring games in which the cheater soloed.)
And even if you break even compared to other people on this site, it makes our statistics un-comparable to those from, for example, diplomacy tournaments, because the majority of the time, the cheater will solo, which means the expected win percentage is lower.
Lastly, it's fair to remove those games, because anyone who thinks they've been cheated can report the game and get it removed.