Thanks for linking to that, that’s interesting! Note he speaks of HM-SMR, and the additional engineering to make that work - code and flash cache layer. For their use case, the 10 to 15% cost advantage is erased by that added complexity, so they don’t go there.
Agreed that DM-SMR is the worst possible SMR type for a NAS. I’m not aware of anything off the shelf that talks to HA-SMR or HM-SMR, though. Which means the shorthand of “your default stance should be “no SMR”” is entirely reasonable.