If it enables massively concurrent in-memory compute then the frequency disadvantage could just be scaled away.

IF you don't care about the issue of moving information, and provided your problem is embarassingly parallel.

Fortunately, a lot of problems are embarassingly parallel.