Wow, this is pretty thorough but I never saw it address the simple question, why not use rdrand?

Also:

> the CPU Jitter random number generator returns roughly 10 kBytes per second on an Intel Core i7

OpenBSD requires at least 8 bytes of random data per TCP connection. Limiting a system to 1000 conn/sec (to say nothing of any other processing those connections may require) probably isn't going to fly.

As a comparison, I can read from /dev/random (and write to /dev/null) at over 300 MB/s.