I’ve measured how R affects propagation delay. From my measurements, it looks like the difference is within 5-10% for R=2 comparing to R=4 (which means that at this point most of the messages are relayed through fast outbound connections), and the distribution of values (50% 75% 99% 100%) is similar for different network groups as well as different R.
Surprisingly, for R=8 75%-99% and 99%-100% delays grow significantly comparing to R=2 and R=4 (about 40% worse). I’m not sure about the nature of this observation.
Another point I’ve found is *network groups themselves does not affect propagation delay much* — at least for a reasonable number of groups (up to n=8, see prior data), which is surprising for me. It may be because of the size of my simulation (only 1000 nodes).
Anyway, from my observations it seems like neither setting R=4 nor using a small number of network groups don’t affect propagation time significantly (setting R=8 does).
@sdaftuar may have more interesting results here.