A sample of 600-900 is very likely to under-represent or even miss the newly registered voters, which is a relatively small group. To capture that sub-population they would need to oversample and then weight, but to do that they would need to apriori identify the newly registered voters and draw a larger sample - neither of which seems to be happening. Hence the buzz does not sound very convincing.
One of several reasons I would like to see Kerry winning by a landslide is to see all those spin doctors being plainly wrong and running with their tails between their legs. I am not above such schadenfreude.
Wojtek