Most polls this time predict a closer outcome than the 2000 results. I haven’t heard a single comparison in all this deluge of media coverage between 2004 polls and 2000 polls. CNN’s 2000 election archive suggests that polls suggested that the race was very close until October 25, 2000; after that Bush gained an advantage (described as ‘7 points’ at one time). Polls predict the popular vote, so this prediction ended up being slightly wrong.
A lead of even 3% translates to thousands of votes; even 1% is a large number of votes in a large state. Kennedy won in 1960 with an advantage of 0.17% in the popular vote, about 110,000 votes; Gore lost in 2000 despite having an advantage of 0.51% in the popular vote (about 450,000 votes).
The polls may say the election will be very close, but they are wrong somehow (which is why there is a margin of error). Most likely it will not really be as close as 2000, and I think it’s possible there will be a clear result by midnight. A scenario where several important states are close enough that the results might be changed by arguing is statistically unlikely. But then, a scenario where several important states have to count absentee and provisional ballots before giving final results is possible.