See Parallel::ForkManager for a convenient way to spawn off new processes and to limit the the number of active ones.

You probably want to extract the host or domain of a link for your commendable desire to rate-limit your requests. URI.pm can do that for you. It's distributed with perl, I believe.

If you find external links in a document, a (perhaps partial) breadth-first traversal strategy will give you something to do while waiting to next hit the current domain. Don't forget about robot.rules. You ought to plan to to honor that.

A CPAN search for Robot or Parallel turns up LWP::Parallel::RobotUA and several other candidates to help with this.