The part of Armstrong's argument I was (explicitly) referring to was not about relative gains of multi-threading but about presumed absolute losses of single-th

The part of Armstrong's argument I was (explicitly) referring to was not about relative gains of multi-threading but about presumed absolute losses of single-thread performance. My argument against this is not refuted by relative gains of multi-threading.. Of course you realize even bigger gains on many common workloads using parallelism, but this part of his argument doesn't need the first part, which was wrong.