Suppose Watson scales cheaply (it doesn't), would it be serious threat to Google? My guess is that it would be but I'd be curious to hear HN's take.
A quick look at the TOP500 supercomputers puts about a 10,000x increase in raw flops since 1993. Using that as a rough benchmark (a very rough one), we're looking at some impressive stuff in the next twenty years. It's not unreasonable to think that:
current consumer processor in FLOPS * 10,000
Naturally, that is a very rough estimate and has no scientific bearing at all.