I have been using Ubuntu + LXD + RaidZ for as long as it was available. Recently (a year ago?) I first had noticed some kind of unclear performance issue with a FLEX container. Heavy file streaming/download would hose and make entire server unusable. I had no idea what to do, and I pretty much randomly converted the LXD storage to BTRFS, and everything was fine. Last few months I ran into same problem with a bunch of PostgreSQL containers doings lots of writes, and same solution helped. Not sure if it's a ZFS problem or LXD, but I will not use ZFS in high load scenario again.

Try temporarily setting sync=disabled ? If that speeds it maybe a fast ssd log device would fix it.

Hmm, interesting, I'll try that, thanks!

ZFS deletes can be really slowed down by having dedupe enabled on the pool.

It's off, but I'll triple check, thank you.

I think deletes in ZFS are just slow, my guess is it is doing work to update/coalesce free segments lists so that future writes stay fast.

Is it slow if you delete a file locally on the NAS instead of via a network file system?

Yes, "rm <file>" locally is what's slow :/