Six 20 TB drives in RAIDZ2 gets you roughly 74 TB usable with solid redundancy, survives two drive failures, and 6-wide is the sweet spot for that RAID level. Not too wide that a rebuild takes forever, not so narrow you're throwing away half your capacity.
64 GB RAM is fine. Forget the old "1 GB per TB" rule, that's cargo-cult math from the Solaris era. For a NAS or bulk storage box you won't come close to needing more, just cap the ARC around 50 GB so the OS has breathing room.
On block size: leave it at the 128K default unless you're running a database on it. If you are, match ZFS recordsize to the DB page size (8K for Postgres, 16K for InnoDB), otherwise every tiny random write drags a massive block along for the ride. For video or backup storage, crank it up to 1M and let compression do its thing with zstd.
Skip deduplication entirely. It sounds appealing but it'll eat your RAM alive and slow everything down. Compression gives you most of the space savings with none of the pain.
Biggest gotcha with 20 TB drives: when one dies, the resilver (rebuild) can run 24–36 hours. RAIDZ2 gives you a second parity drive as insurance during that window - which at this drive size isn't optional, it's the whole point.
