Timing and scalability are probably the biggest issues with realtime P2P applications. Scalability is becoming a bit less of an issue; with mid-splits becoming the norm on cable internet services and 5G SA rolling out, there's just a lot more bandwidth to go around.

Timing issues can be worked around and GPS modules are cheap.

NAT problems aren't usually that bad. Only a minority of networks use symmetric NAT implementations, with most seeming to be using port-restricted cone NAT (EIM/APDF), which can still communicate with any other NAT implementation using endpoint-independent mapping (EIM). Most CGNAT implementations that I've encountered use EIM, with some also doing endpoint-independent filtering (EIF). Between UDP hole-punching, UPnP, NAT-PMP, and IPv6, it's usually possible to establish a P2P connection between 2 endpoints.

A game could use a hybrid client/server and P2P model, with the option to run entirely P2P while accepting its limitations.

A self hosted server executable still works fine with that though. Plenty of games had that model.