This still sidesteps the article's point that what we measure as "efficiency" is itself historically contingent. GPUs succeeded precisely because they exploited massive parallelism within an already compatible model, not because the ecosystem suddenly became open to new paradigms. Your example actually supports the article's argument about selective reinforcement.
There's nothing to agree to disagree about. You're arguing a point the article does not make.