The numbers could be skewed for structural reasons. As in any poll, self-selection is inevitable. The people who agree to speak with a pollster about politics for half an hour are going to be relatively engaged. That partially explains the turnout gap. But it doesn't account for Obama's advantage. (Unless people who voted for McCain are less likely to want to talk about the current state of the country—but that seems like a stretch.) Bias seems implausible. (See the New York Times' methodology here.) Another possible explanation is statistical error, which is plus or minus three points in the NYT and WSJ polls. But again, that doesn't explain the size of the gap.
Chances are, Obama's landslide won't last forever. Retroactive vote reporting tends to be a proxy for popularity. Just ask George W. Bush. In a 2006 NYT poll, more people said they voted for John Kerry in 2004 than voted for Bush. Obama's numbers could drop, too. "Say health care and financial reform hit a wall," says Berinsky. "I think you're going to see some dampening of these numbers."
The takeaway lesson, though, is that asking someone whom they voted for may not be the best predictor of future behavior. Polls tend to ask for voting history in order to provide cross-tabs: How many McCain voters support Obama's health plan? How many Obama voters think he's doing a bad job? But if the numbers of Obama and McCain voters are themselves hazy, then maybe they're a poor barometer for other opinions. A better metric could be party affiliation or whether someone currently supports or opposes Obama. At least that's easier to remember.