Imagine you had magical healing powers. Sitting quietly with someone and holding their hand you could restore them to health. While this would be a wonderful ability to have, it would also be a hard one: any time you spent on something other than healing people would mean unnecessary suffering. How could you justify a pleasant dinner with your family or a relaxing weekend at the beach when that meant more people living in pain?
But you already have these powers. Through the magic of effective charity you can donate money to help people right now. The tradeoff remains: time you give yourself when you could be working means money you don't earn which then can't go to help the people who would most benefit from it.
(I don't think this means you should try for complete selflessness; you need to balance your needs against others'. But the balance should probably be a lot further towards others' than it currently is.)
Update 2012-08-12: this is a response to hearing people offline saying that if they had magical "help other people" powers then they should spend lots of time using them, without having considered that they already have non-magical "help other people" powers.
I also posted this on my blog
No disagreement there, specific arguments ought to be considered. However, in my experience, if someone tells you that you have an obligation to do something (pray to $god, donate to $cause, enlist in $military, vote for $candidate, ...), they are not to be trusted with putting forth arguments, or even estimating prior[itie]s. So, ignore people like that entirely and do your own research from scratch.
So the heuristic is to only consider arguments that don't claim to be leading to any (novel/actionable) conclusions? This rule decides at the bottom line, stopping consideration of arguments that don't conclude with uncertainty or close match to intuitively natural desires, which would be bad if conclusions not of that form turn out to be knowably correct.
If you remain specific, you may get rid of "pray to $god", but not other similar things, "donate to $cause that's known to be worthless", but not other similar things, etc. That should lift most of the load without as many false negatives.