1. The obvious one is “just” extending stuff internally working via Unix domain sockets to TCP sockets. Various internal code is written with an eye to that, including anticipating that certain operations (such as connect) that are instant locally can would-block in a network.

If people enjoy the API, this would be a no-brainer value-add, even if lots of people would scoff and use actual dedicated networking techniques (HTTP, whatever) directly instead.

2. The much more fun and unique idea is using RDMA, “sort of” a networked-SHM type of setup (internally). Hope to get a go-ahead (or contribution, of course) on this.

I mention these in the intro page of the Manual, I think.