I think that Inada is right: there's too much impact on memory consumption and garbage collection to call this an optimization.

Serhiy suggested removing the cache.  But, once you remove the cache, you're going to iterate on all values of all parent classes' dicts to find which values are __dunder__ methods.  This risks being much more expensive for classes with tens or hundreds of namespace entries, making the optimization even less interesting.

So in the end I think I'm going to reject this issue.