With 24 hours to go, the sampling of races I selected from Intrade is pretty much the same:
Patty Murray wins (Washington Senate)
Russ Fiengold loses (Wisconsin Senate)
Brown wins (CA Gov)
Boxer wins (CA Senate)
Buck wins (CO Senate)
Rubio wins (FL Senate)
Miller wins (AK Senate)
Paul wins (KY Senate)
Harry Ried loses (Nevada Senate)
Toomey wins (PA Senate)
Hickenlooper wins (CO Gov)
The only significant difference is that the spread between Patty Murray winning WA State Senate and Dino Rossi winning WA State Senate has narrowed to something like 56-35. That wouldn't be close if it were a poll number, but for an Intrade position, that's almost a toss up.
UPDATE - Intrade correctly predicted 8 of the above 11 outcomes. Harry Reid won in Nevada, Miller lost in Alaska and Ken Buck lost in Colorado. That's a pretty good showing, but I wouldn't call it amazing. A lot of pundits and regular polls did just as well or better. I'll test it again in future elections, but so far I don't see anything extraordinary about the predictive ability of the prediction market.