As I can see we currently don't support document-level parallelisation on `mongodump` / `mongorestore` (and only support collection-level parallelisation via `--numParallelCollections` option) (please correct me if I'm wrong)

Can we consider doing document-level parallelisation on dump / restore operations? I.e. to be able to insert documents for the same namespace (`<DATABASE>.<COLLECTION>`) in multiple (parallel) threads / MongoDB Wire Protocol connections?
(this will allow to significantly speed-up the dump / restore process for MongoDB Server processes with large amount of resources (CPU/RAM/Disk IO/Network IO) available on them)