RE: DeltaBot, that doesn't necessarily support your stance of dropping the units for member counts. That rule (User:DeltaBot/fixClaims/jobs, search for P2124) appears to be designed to eliminate uses of "human" and "person" as units - it ignores more specific units like "woman", "man", etc. It seems a bit odd to have a rule slicing out the more generic units when it would be best done at the constraint.
After quite a bit of research, I can imagine a couple of different test cases that would be problematic for the current member count implementation. Those discussions should probably be had, but probably not on your Talk page.
More practically regarding member counts:
- I'm not entirely certain what programs or scripts have been written to rely on or use the member count property and the "expected" associated unit.
- I'm sure somewhere along the line someone else is going to try and "clean up" the "constraint-breaking" record again - and you'll have to sweep in and remove whatever unit is added each time.
Unless it is absolutely clear that a "less is more" approach is desirable for all known use cases (feel free to point out a particular Wikidata guideline if there is one), I would request that a fairly mild redundancy like the units on member count be acceptable until the associated constraint can be reviewed.