When looking at the screenshots from eval.google.com (see this thread [webmasterworld.com]), I found the second screenshot pretty interesting.
It seems to suggest that Google is working on a database of query refinements. I find it hard to imagine that this effort is just being done for Google Suggest.
I guess refinement suggestions will eventually will pop on results pages, like in "Try adding XYZ to your search".
I also see an incresed datamining-advantage in such a solution compared to Google Suggest, as feedback on refinements is much better here. (With Suggest it's hard to distinguish real refinement queries from cases where people were just too lazy to type.)
A "Try adding XYZ"-feature could fundamentally influence click-through rates on single terms as I see it.
(I hope i'm not circulating yesterday's news here, couldn't that thought elsewhere)