What will happen in Alaska? I think that the early handicapping of the Alaska Senate race, which features primary winner Joe Miller, incumbent Senator/primary loser/write-in candidate Lisa Murkowski, and Democratic nominee Scott McAdams -- is not sufficiently taking into account the inherent potential for random and unstable results. Regular elections between a Republican and a Democrat are fairly predictable from early polls and district history because we can expect that undecided partisans (including those who claim to be independents but regularly behave as if they were partisans) will tend to drift to the candidate from their party. But when a third candidate enters, I suspect things can change, sometimes rapidly.
Here's my understanding of it. In normal elections, most of us use party as a shortcut -- we don't want to bother researching each and every one of the many candidates we're faced with, so we choose instead to make the (logical) guess that party affiliation is a good enough clue to how the candidates will behave in office. And then we move on; we know little of the candidates, and we have neither the easy opportunity nor the interest in learning more. Now, it's also the case that what we "learn" will be biased by our partisanship...strong Democrats are likely to ignore negative information about the Democratic candidate and pay attention to negative information about the Republican, and vice versa. So there are multiple paths that get us to (mostly) party voting.
That can break down, however, in high profile races. The more information we get, the less we're likely to just rely on party as a shortcut -- although in a normal general election contest partisan biases in absorbing information still apply. In a three-candidate, high-information race, we have even more of a possibility of pushing voters away from partisan shortcuts and into choosing based on other information. Moreover, what information we receive is unpredictable, because it depends on the choices of the candidates. Will Murkowski choose to spend the whole campaign attacking Miller? Will Miller return fire? If so, it's possible that voters will hear mainly negative information about those two, and McAdams could benefit. On the other hand, it's possible that one or both of the Republican candidates will choose to run a positive campaign, or perhaps a partisan campaign filled with attacks on Barack Obama (and McAdams). That would seriously change the information mix available. Moreover, it's not clear going in that all three candidates will in fact run fully funded campaigns, but I would guess that it's a whole lot more likely that Alaskan voters will be fully exposed to what all three candidates have to say than Floridians will be to their three candidates. That's also important because multi-candidate contests in first-past-the-post elections are subject to strategic voting effects...a candidate who cannot convince the voters that he has a real chance stands to lose a large chunk of his voters at the end of the campaign.
Add it all up, and once the third candidate is viable (in the sense of having sufficient resources to mount a full-scale campaign) and remains viable, I'm convinced that early polling is much less predictive of the final results than it would be in a conventional two-candidate race. I don't know if there's been any research on this (paging John Sides...I see he's already working the write-in angle), but as far as I can tell it should be the case, and that also would be consistent with my (entirely unsystematic) memory of past multi-candidate elections.