Estimate the time it takes for blocks to be propagated through the majority of the network. Determine the block size that still takes a reasonable amount of time to propagate through the network and use that as the max block size.

Currently, the 4vMB blocks have a median propagation time of 6.5s (or less since the study was done a long time ago). If we can keep the expected stale rates to be <5%, then I consider it an acceptable compromise. Expected stale rate is a function of the time it takes for the block to propagate and the block intervals.