It’s not the mempool, it’s my mempool
Discussion
There still is a theoretical global mempool “the mempool” which is the union of all local mempools.
Given the observer problem, no one entity has knowledge of the global mempool making it a proper probabilistic entropy surface, we only have localized approximations of the global surface.
It’s imperative that no singular observer controls the global mempool policy.
This is an excellent model for a few things I've been thinking about lately, but one in particular is storing historical state in perpetuity. This is very unrelated to your point.
There is an ongoing, and ever increasing, cost to storing the historical state to block 0. It doesn't matter what kind of architecture you have, if you have to store all historical state this is true. This is because, though we pretend it isn't true, our picture of the past degrades as time passes, the organization of historical information degrades due to entropy, and to prevent that you have to have a continuous cost of some kind. The future is foggy, the past is also foggy.
To have a system with minimal ongoing resource investment and no ever increasing one, you have to build a system such that it is not necessary to store all historical data, ideally you don't want to have to store any historical data at all.