I suggest potential users run their own benchmarks. In the handful of tests I ran, I saw *no* reduction in speed in our code. Granted the code is likely IO bound, but take it as one data point.
My biggest complaints have been (1) getting MMS to play nicely with PerlTidy and (2) the need to put "with <role>" after certain methods. There is now a kludge to get PerlTidy to work (see the PerlTidy change log or the PerlTidy subclass referenced here www.openswartz.com/2010/12/19/perltid....
Easy to use and cleans up after itself. Only complaint is that it requires Proc::ProcessTable which is XS and wouldn't compile on a particular platform for me. Seems like checking for a running process require that much overhead.