About 45% of the time the Challenger did better than expected, about 30% of the time there was no break, and about 25% of the time the incumbant did a bit better than expected.
While definitely not as often as I (or most, I think) were under the impression that this occurred, this is still SOMEWHAT significant... it's still a plurality of the time, and nearly twice as often as they break to the incumbent. And if the 3% holds in close elections that's a huge difference.
Plus, if this is due to the undecideds (it was implied in a few other posts that it wasn't, but I'm not sure there's any way to tell for sure... in fact, it seems most likely to me that it IS) the "break" is stronger than it looks... perhaps not in frequency, but in amount... Because only a relatively small fraction of the populace is undecided, that small group swinging the overall total percentage by 3% indicates that they swung heavily in one direction. Of course, that's only IF this effect is due to undecideds... which there's no evidence for or against that.
One thing's for sure, in a close election, this makes it a real b*tch to interpret the polls. Consider the number of EVs in states closer than 3%. If we can't predict which way those states will go, it's impossible to know who'll win. It would be statistically safest to assume that the "break" would go to Kerry... so, he'd win. But with "breaks to the challenger" occurring less than 50% of the time, those are hardly convincing odds!