In 2004 the two most accurate pollsters were TIPP & Pew Research. Both were within half a percent of the actual national popular vote.
TIPP:Obama - 47.2%
McCain - 39.8%
Unsure - 13.0%
Obama - 49%
McCain - 42%
The Pew numbers are a bit older (10/15) however the same margin exists. The 13% undecided in the TIPP poll seems too high this late in the game. Perhaps TIPP accounts for some percentage of weak/leaning support as undecided voters. If that's the case McCain would need to win these undecided voters by a 60% margin to tie the national numbers.
The outliner today seems to be Zogby -- showing a much tighter race. In 2004 Zogby was off by about 3% showing a tied race when in reality George W Bush won by 2.4% national. So based on that data, Zogby underestimated the winner by 3%. 3% + 3% gets us close to the margins TIPP & Pew are showing. Gallup used a traditional likely voter model in 2004 and showed basically the same results as Zogby and the same under estimation of the front runner. -3% Gallup's numbers this year also line-up with Zogby in the traditional model however they now also offer an expanded model that is in line with TIPP & Pew.
So what does this all say? I don't know. I think by all accounts the race is stable however the likely voter models are a huge X factor. No one really knows who is going to show up and vote. I'm sticking with TIPP & Pew for the most accurate numbers because I feel like neither of them have an axe to grind. The big pollsters, being paid by the mainstream media, have less incentive to be accurate and more incentive to be entertaining. For conservatives it's fun to think it's a 3% race. That's almost margin of error. For liberals, it's fun to think it's a 10%+ race. That's a landslide.
With 2 weeks to go I expect the national numbers to stay mostly the same. We've seen no significant movement in the last 2 weeks throughout the debates. There are no more national events where voters can compare candidates side-by-side. The holiday season & baseball playoffs begin to effect the Presidental race in the last 2 weeks. People are less interested in following the day-to-day campaign news. It's less of a topic of conversation.
The only significant change I expect to see would be the last week of polling before the election. This is the point where voters will have mostly made up their minds. It will also include Obama's nationally televised campaign event.