I am a fan of prediction markets as forecasting tools—see this paper on our internal market—but the wisdom of the crowds sure gets elections wrong:
From inevitable to dead and then back to frontrunner.
Markets aggregate information, and that information is mostly polls, and the polls were erroneous. In fact, looking at the market for any given candidate, the pricing is just some amalgam of state polls, national polls, and current buzz. Intrade and other prediction markets are far from perfect, but I suspect the real problem is that elections are simply hard to forecast and thus markets and polls alike are poor prognosticators.
One mitigator would be to combine the two: Run markets where buyers are voters in the given election. The commingling might reduce some of the biases and errors that manifest in the standalone forms and it would constrain market participants to those with more information than the average person—say, folks actually caucusing. One downside, however, is that participants would have a pecuniary incentive to not change their minds.
As for the current pricing of a Senator Clinton victory, I suspect the markets (as with the media) are overreacting to her New Hampshire comeback, overcompensating for their earlier pessimism. I would put the odds at 6-to-5 and pick 'em.