What is the theoretical highest ELO chess rating possible?  Someone might cross 3000 in chess.com, but how much higher could it go (barring any bans for cheating :-P )?

The sky is the limit. Theoretically there is no Elo roof.  

Chess.com actually uses the Glicko system for chess ratings, not the Elo system. Neither system has a theoretical maximum rating. However, there is a practical maximum in that the higher your rating is above your opponent, the less you get for winning. The highest rated player can only be so far above the next highest rated player, and they have to be able to beat the next highest rated player consistently to get very far past them.

The theoretical maximum for a given pool of players, then, would be the number of players in the pool multiplied by this practical maximum, yes?

 Well, the practical maximum I was talking about was sort of the theoretical maximum for a given pool of players. The practical maximum between two players would be the rating difference at which the higher rated player nothing for winning. That number times the number of players minus one would be the theoretical maximum range of ratings for the pool. But evertime a rating goes up, a rating goes down, so the theoretical maximum rating for a given pool would be half the the theorteical maximum range plus the average rating (which should be near the starting rating).

However, at that point you have a uniform distribution of players. IIRC, the Bayesian statistics of the rating systems assumes a logistic prior distribution of the players, so the use of that rating system on a uniform population is questionable.

My understanding, though, was that the worst player in the pool, the one rated zero, would never have his rating reduced to less than zero so the axiom that for every rating point gained there is a rating point lost doesn't hold and you could in fact arrive at a uniform distribution with each player separated by the maximum rating separation via this theoretical source of inflation.

The fact that there would be a player rated zero, however, does mean that my original assessment of the maximum rating needs to be revised from MAX = S x N where S is the maximum rating separation between two consecutively ranked players and where N is the number of players in the pool, to MAX = S x (N-1).

There is no minimum rating in the Glicko system either, AFAIK. It is theoretically possible to have a negative Glicko rating.

Also, there is no axiom in the Glicko system that for every point gained a point is lost. That's why I said the average should be near the starting rating. It generally balances out, but there is no mathematical constraint that forces it to do so.

I see -- the rating floor of zero must be a misconception of mine.  I stand corrected.

Enlightened very much

Thanks every one commented

My father searching for chess rating above2400

Would someone like to test the minimum limit theory, guinea-pig-style, lose spectacularly to a string of wood-pushers, and force their rating below zero?  I'm curious about this, but I'd hate to ruin my current rating, and you're not allowed to open two accounts at chess.com :-)

http://www.chess.com/echess/profile/T-95

i got a long way to go, i'm 900!

my calculation shows the highest rating is 4100.the average steps must reach around 286 steps.