[mnutt]

Are these sorts of things latency-sensitive enough that they couldn't use them with EC2 and similar? Or are the bandwidth needs so large that bandwidth charges cancel out any compute savings?