I would assume, that the cost for parallelization in Python is memory, to spin up more runtimes, VMs, instances or whatever, to enable the scale-up. However, not a single mention of RAM usage, visible at quick glance, on the whole page? Seems like only the good parts are presented.