I've used fork in a somewhat more specialized version of my Fork::Queue node. I'm using it in an error event processor. Basically, I don't want the clients that are sending events to have to wait on the server to process the event.

I fork off one child, who can work the queue of events while the main process is free to handle client requests as fast as possible.

I also use fork() when I need simple multitasking, and don't want to serialize file i/o ( or some other long-running or blocking operation).

As an example, I have a unix box with 3 web servers on it. I have a perl script that grabs lines from each web server log and grinds out some stats. So, the parent forks off three children, each with a pipe back to the parent. Each child process grinds on the individual stats, then passes a small amount of aggregate stats back to the parent over the pipe. The parent can then spit out "global" stats for the box. This runs, as you can imagine, almost three times faster than doing it serially.