I never post on here anymore, but I do occasionally read my friends page... at some point, someone said a while back that the presidential race is only close when viewed through the lens of the media, who have staggering financial incentives to portray it as such (advertising dollars).
Looks like this is rather well supported by a group of CS and poli sci students at one of my almae matres, who put up this site forecasting the outcome of the presidential election based not on popular vote, but Bayesian estimators of Electoral College outcomes based on continuously updated state-by-state poll results.
You can select various political climate scenarios, from "Strong Democratic" to "Strong Republican," which artificially slants independent voters +/- 10% for either Obama or McCain.
At present, under the Neutral scenario, the model predicts a 1.00 probability that Obama would win, if the election were held today. Under the Strong Republican scenario, that probability drops to... 0.99.
EDIT: Going back through the archives shows that McCain did get a major post-convention bump, and was the predicted winner the week of 9/15 - 9/19. But the overall forecast history from 7/31 to today show Obama winning handily.