Basically that's what you're saying when you characterize the results as produced by "outraged nuts [who] generate more web space than a lot of other people". But again it presupposes your biased view that the nuts are all on one side. Normalize for symmetry in nuttiness, and you still have a higher volume of email smears provoking such nuttiness.

Very interesting suggestion about restricting the search to sites of known relevance to the domain of the query. Remember, though, that I built the Oracle not to explore the narrow question of Republians versus Democrat, but to examine a wide spectrum of "polarized" questions. Restricting the search to specific sites would require specifying a list of such sites for each search set. Not to mention building my own Google search spider. Yikes!

A secondary problem might arise from the choice of sites. If they're news sites, the results can be skewed by political bias in their coverage. If activist sites, the problem is exacerbated by the relative intensity and differing styles. whitehouse.org is primarily a satire site with a bite, while moveon.org is pure activism. I'd think the results would be qualitatively different.

I'll have to ponder the question some more. Potentially it's a really good idea, but I don't know a practical way to implement it.