We're trying to see what is the worst performance in terms of I/O, i.e. when the database just started up or the data/indexes being queried are not cached at all.
You could create new tables that are copies of the existing ones (CREATE TABLE foo as SELECT * FROM ...), create new indexes, and run a query on those. Use schemas and search_path to keep the queries the same. No restart needed! (just potentially lots of I/O, time, and disk space :) Don't forget to do explain (analyze, buffers) to double check things.
