First off, let me just say that I've never actually done anything resembling this in the "real world". So take my advice with a grain of salt =]. (Or two.. or three)

If I were in your shoes however, I'd simply test a couple of different hardware configurations. Find the main, generic, thing that your application does, and devise some way to benchmark it. In this particular case it's some sort of webapp, so your "main, generic, thing" is responding to user queries. So something like ab (apache benchmark tool) will probably be very suitable for testing your application.

Next step is to find some hard ware. You probably want several different "levels" of hardware. For example, you might test on a 286 with 32 megs of ram, a 1ghz with 256 megs of ram, a 3ghz with a gig of ram and then something like a quad 3ghz machine with a shit load of ram. Now compile statistics for how fast your app performs on each platform and compare these statistics to what your customer expects.

Perhaps you find that your app can handle 30 hits a second on the 1ghz machine, but the customer wants to handle 40, so you suggest a 1.5ghz machine. Or something along those lines. I've probably over simplified the problem, but this might be a decent approach.

As for which brands, well, unless you want to do some *major* testing, all you can really recommend is what you, and your company, has discovered through actual experience. You might want to combine your experience with "reputable" magazines, for example "we use Foobar's hardware here, and it's very reliable, x,y and z also recomend Bazfoo's hardware, but we've never tried it".