Unless I've misunderstood exactly what you're referring to here, there is also a lot of C++98-style SFINAE which relies on ellipsis conversion being bottom of the overload resolution priority list. While you wouldn't write code like that today, if this paper were to deprecate single-ellipsis functions like f(...) it may be more of a breaking change than anticipated to people who are still using some rather old libraries which rely on those techniques.

Glad you're not deprecating that currently, but food for thought should the discussion around this proposal turn to whether it's worth bundling f(...) in too.