The top-level task didn't run successfully after this was deployed
due to the worker being killed due to heavy disk usage. While the
more parallel version does log much more, it doesn't totally explain
the disk behaviour. Nonetheless, reverting it is sensible to give us
the time we need to investigate more.