As the average difficulty goes up, pools tend to run mining software that have a predictable reorg policy, in order to minimize the probability that their own blocks get invalidated. That is why we don't see large reorgs these past few years.
Could you elaborate on this? What is a reorg policy, and are there unpredictable variants?
I would have guessed that the decline in reorg frequency and depth is the result of lower miner inter-latency.