Like Mark, the incumbent rule was the big reason that I thought John Kerry had a good chance to win in 2004. But it didn't turn out that way. Bush was ahead by a sliver but below 50%. And he ended up being up by a sliver but over 50% on election day. Boy, was that fun.
See Mark's look at the academic literature.
Where the incumbent rule has always been most powerful is in congressional elections and those lower down the ballot. In general you know the incumbent far better than the challenger. If he or she hasn't made their case by election day the 'undecideds' are probably disproportionately people who've decided against them but hasn't fully admitted that to themselves. And if they've already decided not to vote for the person they know, more likely to take a chance on someone who might still prove themselves if given the chance.
But that logic doesn't hold as much in a presidential election, certainly not in the current climate of a year long pervasive media saturation. At this point you've heard a lot about Barack Obama and Mitt Romney. A whole lot. This after all has been the premise of the Obama campaign and the Bush 2004 campaign for that matter -- turning the election from a referendum on the incumbent into a choice election.
As Mark notes, the incumbent rule seems dormant in presidential elections. And, critically, in key states Obama appears to be at or near 50%, which means that if the polls are accurate. But we'll know soon enough.