Maybe you could get better control of the hung processes if you ran them as forked children?

On a completely different tack, perhaps you should go high level and use LWP instead of IO::Socket to see if HTTP/S communication is possible. Might make your problem much simpler to manage.