I generally wouldn't suggest using back of the envelope calculations for much of anything, by the way. Dropbox has stupidly attractive unit economics and a proven customer acquisition channel which explodes like whoa. There. That is probably accurate and possibly useful. Dropbox has N% margins, as demonstrated by back-of-envelope calculations? That is perhaps not quite so accurate or useful.
Back of the envelope calculations aren't necessarily useful, but they're fun. They let you analyze problems and try and think of all the different moving parts that go into building something - that's why they're so popular in hacker circles. The conclusions are probably useless [1], but the debate and analysis is what you take away.
[1] I can't remember what it's called, but I remember reading a theory (on Wikipedia) that says that people tend to estimate pretty well when they estimate the multiplication of lots of numbers. Mostly because, on average, they make mistakes in both directions (i.e. half the numbers are too high, half are too low).