I've written pretty extensively on search[1].. Basically, I've come to the conclusion that there are opportunities to improve search, and I've even gone so far as to start a company around it:
The problem is it has to be niche to compete with something like Google. For instance, my project targets financials (and might have info you're interested in).
What you're mentioning related to decentralized search is difficult.
Distributing something like search would be hard because of the aggregation factor. Google, my system, DuckDuckGo, etc. all require some sort of searchable graph (I actually use relations, but more on that later). To search a graph on a distributed system is fine, but aggregating and redistributing the information in real-time is very very difficult. Aggregation typically requires one node being the final source of truth, which then redistributes that information.
I suppose it'd be possible if the search results could be delayed. Perhaps for a wikipedia type of project, where it's not the links that change, so much as the content.