I still think that the argument is flawed beyond trivially parallel problems, but my understanding is that the author is arguing for shared-nothing, not for single threaded/single-process solutions.