I think the relevant axis may be short-term/specific vs. long-term/broader consequences rather than unimportant vs. important. I think defecting is usually a long-term bad strategy but a short-term good one. If you're pretty sure there's not going to be a long-term unless you fix your short-term problems immediately, defecting might be a good idea for you or your chosen cause - not sure about for the world at large.
Scott, known on LessWrong as Yvain, recently wrote a post complaining about an inaccurate rape statistic.
Arthur Chu, who is notable for winning money on Jeopardy recently, argued against Scott's stance that we should be honest in arguments in a comment thread on Jeff Kaufman's Facebook profile, which can be read here.
Scott just responded here, with a number of points relevant to the topic of rationalist communities.
I am interested in what LW thinks of this.
Obviously, at some point being polite in our arguments is silly. I'd be interested in people's opinions of how dire the real world consequences have to be before it's worthwhile debating dishonestly.