has anyone ever done the modeling on the dynamic limits of decentralized network consensus? my intuition has always been that getting, say, 100,000 globally distributed voting nodes to agree on even a simple truth value will quickly run into exponential (or worse) latency bottlenecks by way of metcalfe's law.

  the speed of light, as a hard limit, starts to become insurmountable when your consensus pathway has to be traversed  the equivalent of tens of millions of  km of signal path in order for every node to have an equal influence on the validity of every other node's vote.  

these high tps coins all seem to get around this by weakening decentralization, whether by shortening this path, either because they have far fewer nodes, or are deciding to sum over some approximation of node consensus by sampling or creating privileged paths, or because geography already creates privileged paths (all the nodes are in the same general area, or even in the same datacenter).

I suspect the reason TX speed tends to stay mired in the tens per second region is that no amount of consensus accounting gimmickry will overcome the underlying physical limitations of true global scale decentralization of consensus.