For anyone building something similar, I imagine having an elastic load balancer for TCP port 22 with a health check on a web service that spawns a process as each of the user accounts before returning "good", combined with an auto-scaling group to make certain there are always a couple healthy instances, would be an automated way to keep something like this running through fork bombs.