I implemented a similar system at $work using PoCo::PreforkDispatch. It works pretty well and does replace crashed workers. If you are concerned about frozen jobs, you can easily avoid this problem using a classic alarm signal.

We're having just one issue with this setup however, sometimes the controlling process gets completely stuck on something and has to be killed. I don't know yet if PreforkDispatch is the culprit, but I doubt so. I can keep you posted if you want. Update: this is a MySQL issue, POE and its components have nothing to do with that.

(updated) As for father-children task/result IPC, there is no need for XMLRPC. I think this piece of code is a left-over of an earlier design. Actually, PoCo::PreforkDispatch leverages POE::Filter::Reference, which will serialize using Storable the arguments and return values of your function.