I can't talk about the largest production database, but our largest internal cluster we've tested on so far is 128 machines and hit each one with our own continuous load generators and a chaos monkey (who turns on and off machines at random).