Really? Given what Notch plans (2,000 DCPUs per machine - which is certainly doable), let's assume he's running a high-cpu instance (large) at Amazon for those 2,000 DCPUs. That's $500 per month for the machine. If he can't get that much money out of 2,000 players, he's doing it wrong.
Assuming you allow for eventual instead of atomic consistency, and allow for loss of CPU for several seconds if an instance goes kablooey, the load distribution is really not that hard. (And your spaceship really doesn't need five 9's. It's a game, so just blame it on "an ion storm". Or "a space monster" if that floats your boat ;)
Partitioning does not need to be too clever if you can constrain the number of ships per region - certainly something you can impose via game design.
Lag only matters if you interact twitch based. Notch's plans lend themselves to indirect agency via the simulated ship's computer instead, so if everybody perceives the world as it was a few seconds ago, not much is lost.
It's certainly a doable task. Notch's mind is in the right place to pull it off. No doubt, it'll be hard - but it'll be interesting to see what he comes up with.