This isn't surprising, the simplicity of the problem does not exceed the overhead of parallelism. A task like that is actually quite rare in practice, but easy to stumble upon when doing demo or example code.

Even with Erlang I don't expect much improvement, when what you're solving is trivial for a single core, adding more cores and processes does not help.

Yes, mainly because pmap is an early demo of how one could do parallelism in Clojure. It remains useful for mapping very heavyweight functions across relatively small data sets. It is not a substitute for Fork/Join, which is on the way.