I don't think the controller needs to be a shell script, in fact, I think that in a perl script it might be easier to control the memory usage (just a hunch). I think that as long as you don't create and destroy loads of lexicals in the controller, you should be alright.

Ofcourse, if you have a simple controller, it should be easy enough to test for leaks - just do a couple of million iterations though the controller's mainloop without running the childs.