We mirror our customer's hosted databases in high performance mode. About 140 databases, from one server to a twin located on the same network switch. The purpose is simply to provide a very warm backup in case of hardware failure - much cheaper than clustering, and everything is redundant - not just the front end of the machine. The limitation is simply the number of worker threads on the mirrored machine. 5 workers per DB on the mirror, 2 workers per DB on the primary. On a 4 proc 64 bit machine the default number of worker threads is 512. We upped it to 1000 and have observed no problems at all. The mirror has very low CPU usage, the network traffic is minimal.