Pike's definition was "handling many things at once" vs "doing many things at once"; Sawzall or, to take a simple example he also used, a dot product where you compute sub-products in parallel, is basically doing one logical thing at a time - you have your inputs and you're computing the output. You're doing it with more resources but it's one thing. SIMD instructions or superscalar execution is like that and everybody calls it a form of parallelism and nobody ever calls this concurrency, though the Wikipedia quote could be used to defend calling this concurrency.

I think I don't want to argue about whether parallelism is a special case or not, really; I'm fine with it being a special case, as long as I get the point across that this special case deserves special treatment and is underserved by tools aiming at the general case...