Private information retrieval (PIR) protocols assume multiple parties have copies of a search database. How practical is this in real world? (In more details the 'practical' term is: Will Google be ready to allow other independent search tools copy its database for such PIR? If Google allows, then I assume that there should be no loss in its business)
Information theoretic PIR protocols that have communication complexity less than the size of the database require multiple, non-cooperating servers . In a computational setting, this is not required. So if you do not want to have to share your database with multiple servers, you should probably use a computational PIR protocol.