A small percentage of voters -- 2.5 percent of those surveyed -- said they may turn away from Obama because of his race.The article claims that Obama will lose 6 percent due to his race, but I find no mention of whether or not they asked if anyone was voting for him because of his race.
The concern I think this article is implicitly trying to get at is what is more popularly known as the Bradley Effect or the Wilder Effect, which is the observation that sometimes black candidates do systematically worse in actual elections than they do in the polls leading up. My understanding is that the academic literature has not found this to be the case (or has disappeared), and that it is more likely a flaw in poll methodology.
For those who are more interested in that subject, I suggest they turn to this EconTalk interview with polling expert Doug Rivers and tune in a little after the 40 minute mark.
P.S. A while back the DOL was throwing out predictions for the outcome of the election. I'm betting a Democratic sweep, as both Congress and the White House go to the challengers. Note that I prefer a split, and I don't care which party gets which branch.