It's probably not a big help at the moment, but if your friends had mock pieces of the application for testing, they could isolate each part in turn and do benchmarking that way.

For example, when testing something like a libnet module, it's a whole lot easier to whip up a fake Socket module that does exactly what you want than it is to work with all sorts of network configurations. So they could have a fake database object, a fake web server object, and so forth. The performance isn't quite real-world, but if they have enough fake pieces, they can benchmark each real piece in isolation. That can be handy.

Hey, testing can't solve everything, but this is a place it can help.