Ok, yes, I admit that it's possible to benchmark this stuff but I do that only with the caveat that the alternatives have to take into account number of non-regexp opcodes dispatched by the /foo/ or /bar/ and what version of perl you're on, etc. I expect it to be an especially fiddly bit of business and I don't trust it unless it comes complete with a perlguts explanation of each part that's being used or not used.