I think that is an excellent vision but to understand the challenge of achieving it consider how a search engine has to operate. Typically it 'shards' the index of all available documents into chunks, where the necessary descriptive bits for a chunk can fit in the memory of a dense memory server. Then each server returns its "best" result which are collected, and from there the top 10 are selected.
Search needs to be fast. When you click search you want the page to then pop up with your 10 blue links. The physics of light suggest that your index files all need to be close together, both physically and with high bandwidth to support a couple of round trips for distributing the query and returning the results.
While I learned how search worked while I worked at Google, it was a brilliant engineer named Keith Peterson at Blekko (now at Google :-) who taught me some of the more subtle issues involved in balancing indexes, shards, and query processing in order to maximize utilization.
That requirement of closeness and high bandwidth, makes federation hard. Even forwarding a search to Bing (via the Yahoo APIs) to check for additional "long tail" results really challenges your ability to deliver fast results.
Unlike CMSes where they could run independently in parallel, search is a 'talk to everyone all the time' sort of problem, you don't know apriori which indices may have relevant documents (which is why it is called "search" ;-))