There is no reason to run the conversion to documents not in a parallel stream. This should speed things up in theory because the buildDocument call can be slow for large documents. It will however use more RAM so probably need to feature toggle it.

Also need to verify that it is indeed faster.