Doesn't this depend on the number of servers, crash rates and recovery times? I wouldn't feel confident running 3 servers running at 80% capacity in ultra low latency scenarios. A single crash would overwhelm the other 2 servers in no time.