Curious now - were you running an unconventional workload that stressed LevelDB, or do you think some version of this advice could be applicable to typical workloads?