Not strictly connected to your tests, but:
As of ZFS, we've had experience that it degrades over time after random updates because of files becoming non-linear and sequential reads becomes random.
Also there are Q about ZFS block size - setting it to 8K makes first problem worse, setting it to higher values means that 8K write will need a read to recreate the whole block in new place.
Best regards,
Vitalii Tymchyshyn
