Yesterday at our Toronto Transhumanist Association event I was debating the Simulation Argument with local physicist Matt Schultz. We were discussing the processing power required to run a sufficiently complex simulation and I noted that even the most sophisticated simulation could be expressed simply by slowing down the clock speed to give the computer time to crunch the numbers. Schultz complained that at some point a massively reduced clock speed would defeat the purpose of the simulation for those running it.
True, that's an interesting limitation. If you're running a simulation for the purpose of observing the results, and if you have to slow it down so much that it runs at an unacceptably slow rate, it would simply not be worth it. That being said, if the simulation is for the benefit of the simulated agents or participants inside the simulation (where clock speed is irrelevant -- time elapsement will still feel "real time"), then the external clock speed of the supercomputer is a non-issue.
Moreover, with the potential for quantum computing, the whole issue of processing power may be a moot point anyway. But as Schultz told me, "Simulating everything with qubits might indeed make it a moot point, but that's assuming a way can be found to actually stabilize that many qubits. Keep in mind that qubits are formed by entangling every subatomic particle in the system, and then protecting them from any outside disturbance until the compution is complete and computers' wavefunction can be collapsed. Every qubit added to the system makes the whole thing essentially exponentially more difficult to maintain. Not that I'm saying it's impossible, simply far more difficult than anything we could possibly contemplate today."