With fine-grained parallelism, programs don't tend to start new threads per bits...

With fine-grained parallelism, programs don't tend to start new threads per bits of work. Instead there are a bunch of worker threads that get work delegated to them in various ways. It's this way in parallelism libs for all languages.

The Clojure example uses pmap, which also uses an existing pool of threads. (source to pmap shows that it's backed by futures: https://github.com/clojure/clojure/blob/clojure-1.10.0/src/c... - futures are amply documented to be backed by thread pool)