Latency matters less when it’s all run locally and can even be done predictably with local ai for many scenarios. That with local application rules that get verified by whoever is peering through/with you on a app.
Another part that I came across as a separate inquiry of possibility was kind of a OS on top of such a structure where you wouldn’t need anything than a “terminal” as all your computers are on a virtual machine overlay. It’s be interested to have every possible computation be hashed so once it’s done it just need to be retrieved so you can treat all peers as a type of addressable memory to not only information but computation and data.
It’s something I wanted to build but then I lost motivation because I can’t possibly be sane or unique now can I?