What we have here is a basic serious flaw in the polls.
Some polls just contact their likely voter sample, and adjust for demographics (race, gender, etc) before posting results.
Some polls contact likely voters, then look at the parties. If there are too many Republicans, they count that as a demographic, adjust, and post.
The polls that say "Republican" is a demographic show a tight race. The polls that believe that people could change parties and therefore do not adjust for party affiliation are showing a large Bush lead, coasting to an easy landslide victory.
Which is correct? We have no clue, because this has never happened to this degree before. In elections past, the parties in the poll closely matched the ratio that voted last election. However, this year we are getting a ton of Republicans answering the phone, so some polls are treating that as an error to adjust down to a tight race.
The basic question is: is the republican sample an error, or have people actually shifted Republican? We wont know till the exit polls come out. After that, hopefully polls will be more accurate in '08