^ Yup. This confusion between majority-of-ordinals and OBO was cemented by a New York Times article shortly after the event which put U.S. judge Joe Inman on the hot seat.
Under the majority-of-ordinals system (used, for instance, at U.S. nationals), if one judge (Inman) had switched his vote to Kwan for second and Slutskaya for third, then Michelle would have won. This would have given Michelle a 5-4 majority of first and second place ordinals combined, and Michelle would have been second in the free and the gold medalist overall (as Meli Huber posts above, #3).
But under one-by-one scoring, which was used in international ISU championships at the time, the analysis by Buttercup (post 4) is correct.
Two judges would have had to have changed their minds between Kawn and Slutskaya for Michelle to have pulled it out.
The short program was another story, because only Kwan and Slutskaya were in the running for first. In the case of only two skaters, the two systems are the same -- majority of first place otrdinals wins, nothing else matters. Irina fans who think that their skater was robbed point to the SP, where indeed, a single change of vote would have made Irina the SP winner (as mskater points out).
BTW, people who analyze voting systems have actually proved mathematically that no system can possible work perfectly in the case of three or more candidates. For instance, "Condorcet's paradox," where A beats B, B beats C, and C beats A, can be avoided only by awkward distortions elsewhere in the system.
The CoP was supposed to speak to this problem (but then Lysacek and Weir exactly tied at U.S. nationals, so there you go.)
In '06, with Kwan sadly out, I was pulling for Shizuka Arakawa, Irina Slutskaya, and Fumie Suguri to make the podium. 
