I think one of the overlooked things of the process-per-request CGI scripts we ran back in the day is how you might run out of PIDs. I remember this happening once, but I think(50/50) it was because the PID was a 16-bit value at the time and the system and an uptime of about 6 months.
PIDs will recycle rather than exhausting. As an aside, that's also the reason why it's unsafe to use a PID from any process other than the parent process. As soon as the parent calls wait(2), the PID is released and could be reused for a fresh process at any moment.
The only way to run out of PIDs is not wait() properly, leading to zombie processes, or else some CGI processes time out and don't exit, so never reaped by a wait().
