"Sorry, there were no results for your query: which tennisplayer beat andre agassi"
that is pretty awful and funny though. it can do all that but not see that tennis player and tennisplayer is the same thing :)
That said, scaling it isn't trivial and I don't want to sound overly dismissive. :)
I firmly believe that we can search the whole web this way, but Powerset doesn't have the hardware assets at the time I write this post to do the whole web (or we would have done it).
It's never going to be the case that the PARC algorithm we use will be as fast as Google's algorithm. There is simply a lot more work to do, so they stay a fixed width apart.
But the way that the industry looks at computing resources has changed a lot since when Google is founded. They're becoming a commodity to the point where you can rent CPUs on a hourly basis. Also, there are new kinds of hardware that are starting to be tapped (for example, over the last few years graphics cards have become more and more a platform for general parallelized computation) that make the task seem feasible.
However if you really use it the right way, asking questions of the type "which tennisplayer beat andre agassi" which is an example I saw somewhere then you see the power.
I am hardly an expert but I guess the problem is that the monetizeable part of search is the easiest search, keyword search. And beating Google there, well good luck...
And the search that Powerset does so well isn't as easy to monetarize.
So as someone said, the best way for them to make money might be through licensing.
I guess there wont be one form of search to rule them all(nor have there ever been) but many different ways of search coexisting and cooperating.
One thing I wonder is, is the scalability a money-issue, ie can they buy truckloads of servers and index the whole web with their technique or could their algorhitms simply be too complex to scale, ie NP-complete( http://en.wikipedia.org/wiki/Computational_complexity , http://en.wikipedia.org/wiki/NP-complete )?