The documentation for forkperlipc, as pointed out by McA below, has an example of using a hash to store all of the PIDs that have been stared. You may want to study that example.

So, you can think of it this way. A global scalar variable in the parent is like a bucket that can only hold one thing. Every time that you assign to that variable, perl dumps what is currently in the bucket, and puts the new thing into the bucket. Therefore, only the last child PID will be available to your parent process, at least in the way you currently have it implemented.

A hash is like a pegboard, where you can hang many things off from the single hash variable. It differs from the bucket analogy above by storing the pegboard in the variable instead of storing a thing in the variable.

note: not happy with the analogy - best available between my ears at the moment.