I have to admit that I myself rarely need to use it. When I first read about FizzBuzz I tried to solve it by division and then comparing the floating point result to a rounded int result (i.e. circular way to get the remainder).
I'd be interested to see how many people pass if you simply remove the remainder operator part of it entirely. My guess is that near 100%.
FizzBuzz is still less "unfair" than a lot of the trick programming questions, riddles, or random CompSci trivia that many companies (Google, Microsoft, etc) do.
This is one of the reasons I would never even apply at a Google. I don't perform well programming wise under that level of pressure and my "random trivia" reserves are low.
Unless they "cheat" this should give you a good idea of what they're capable of. The cheating should also show up if they cannot explain it or why they wrote it the way they did.
Additionally, they could spend some time with the team they'd be working in, pairing (or whatever) with other members and really demonstrating their abilities.
It's not really that difficult to get away from the whole "technical question at the interview" problem, and I'd wager that in the long-term it'd be better for all concerned.
I'm sympathetic to complaints about overly technical interviews (I've bombed a few in my day because of stress and/or ignorance), but as others have pointed out, this is FizzBuzz. I'm generally very bad at interview/algorithm puzzles, but I could even figure this one out pretty quickly.
The big difference between the two metaphors is that mine has direct impact on your personal life and negative consequences are gonna be seen as your fault, while yours is a company threat - fixing it will get you seen as a hero, not a disgraced villain.
Even Google lets you try again in a year.