Which is why I think we need to look to another web/network, which is people's social connections.
We just need a way for everyone to maintain and share a personal index that contains both information from automated scraping (i.e. keywords, SIPs, etc) and any notes or other metadata the person cares to attach to the page, domain, etc.
Of course, an index large enough to serve most queries you might be interested in would be pretty big, similar to the size of the document set.
Inverting the Web
The inverted index of the documents themselves is the biggest part. And since it's easy to verify, perhaps that part can be handled in a more centeralized, possibly federated, fashion?