Interesting - it could be due to how Spanner implements it's backend - if each database is a separate file, that could be more efficient to search through? (Maybe at the cost of opening a file for each db)