For the purpose of constructing a financial model, I've been asked a few times (over the course of years) to provide an estimate for the variable costs involved in hosting, or more generally, keeping a web application online.

I'm curious to hear various approaches people have used, from wild-ass guess approaches with a few basic assumptions, to more rigorous approaches involving benchmarking.

Note that the technology would not be built yet, so true load testing isn't a viable option.