When do you ignore your users?

Quoted in Getting Real is a passage by Jef Raskin about avoiding "feature blight" (http://jef.raskincenter.org/unpublished/widgets_of_the_week....). The passage suggests too many features is a bad thing and that the best way to avoid that pitfall is a very real and constricting deadline. My question is that once you've launched and you're able to afford a more liberal deadline, how do you keep feature blight from creeping in?

37 Signals was noted here recently for refusing to implement suggestions until there's a critical mass of users looking for them, but is that the "best" way to combat this issue?