I would suggest to run a simulation where nodes are globally distributed and have various latencies with high variance, and then study this problem of non-uniform distribution of information. with GHOST nodes would cluster in physical locations (say Iceland). the latency between those heavy nodes would be very low, and the latency to say Australia very high. the heavy nodes could then co-opt the network. it does not help if the authors proves a bound of latency, because this imbalance would destroy the network. with geographically skewed distribution of information, one can imagine all kinds of weirds effects and attacks. perhaps I'm wrong and will find out this actually works, but even then robustness requires safety over the longterm in very unexpected cases.
The information distribution should not be any more skewed than Bitcoin because every block can be validated by itself assuming only the validity of its parent as prior knowledge. This differs from the Israeli authors' GHOST, which assumes that nodes already have the uncles; in our system, uncles are included in the block. But we are definitely going to be doing various kinds of network simulations to make sure that every network protocol we push out is stable and convergent.
