Reloading those tokens takes around the same effort as processing them in the first place.

It's ok to be ignorant of how the infrastructure for LLMs work, just don't be proud of it.