You might benefit from the same technique for a "special" vdev... If you run an all spinners array, but the metadata is on a small partition, you might get better seek times during metadata scans, since everything will be closer together? And/or if you use raidz for your regular vdev, but mirror for the metadata, you'll have more capacity for metadata reads.
Your NAS sounds pretty slow, it shouldn't take an hour to scrub 1tb of data (on average)
I would be in a different world if my system was backing banking or other loss critical data. For all my workloads, losing 5 seconds of writes in a crash is NBD.