If it doesn't work, it won't be because of technological impossibility; it will be because of bad business strategies such as increasing their user base faster than they can lease and implement hardware.
Even Slashdot, normally full of intelligent people, has posts like this one which ends in:
Anyone who believes this is technically feasible, much less economically viable, is fucking *retarded*.And he got modded +5 Insightful! (This post combined with people's input lag fears make up the majority of naysayers, sprinkled with a few people who are opposed to not actually having the game on their computer.)
Some people are right: 100 ms does matter when playing a multiplayer FPS. But even that is offset by the fact that, between OnLive clients, lag is all but eliminated.
Also, it's much easier for humans to account for visible lag (they press a button, wait 100 ms, and then a shot is fired), which OnLive will give us, than invisible lag (they press a button, a shot is fired, and 100 ms later their enemy may or may not have been hit), one of the current norms in multiplayer FPSs (there are ways around this). Previously, weapons like Quake 3's machine gun made sound effects that helped the player figure out whether they were actually hitting their target.
I think this has an excellent chance of succeeding if the price is right. OnLive probably won't be perfect even if it performs exactly as described, but it looks right for me.
I highly encourage people to watch the 48-minute demonstration video. I, for one, welcome our new cloud-gaming overlords.
OnLive is expected to be released "Winter 2009" which, I guess, includes the first few months of 2010.