Based on personal experience as an enterprise near-line NAS developer for a well known company - you may be using our laptop - serialisation doesn’t scale well.
It’s easy to code, naturally, but if you’re saving large objects your performance will tail off over time as the volume of data increases.
Wins I’ve seen and coded myself have been in reducing the iops.
I mean yes, saving more to disk is slower than not saving it, computing more is slower than not computing anything... Or are you making a more subtle point?
The primitives developed for the system are incredibly flexible and potentially worth knowing about.