I hear the same performance happens with our without the session cache enabled (for benchmarks). The http(s) benchmarking tools don't resume sessions. It's simulating a horde of new clients who never come back or request other resources.

It would be interesting to see stud with a session cache too.