In my view, (Moose is) too big, too slow, and does too much for its average/median use case.

I agree, and use Moo by preference wherever possible, including in some DBIx::Class result classes. However, the OP is all-in with Moose, I believe.

(Some people would make the same statement about DBIx::Class. There is a lot of careful tooling required to effectively use it even without adding the Moose layer. Most applications don't warrant the extended effort required to make it work well, so using it in those applications can easily become an exercise in frustration. For larger frameworks it becomes indispensable, as does Moose in certain cases of its own.)