Not an easy one to answer, but let me start by correcting the 150ms, this got out somehow a few months back when we were in the infancy of deploying CPPM in VPC's / VNets.... 200ms RTT like what is used for traditional CPPM on-prem clusters, but note this is a guideline, RTT is only ONE SMALL metric in relation to clustering over a WAN, the TRUE figure is data-path throughput, and that's very much tied into my answer below that there is a correlation to load, and load is the sum of authN, Captive-Portal, TACACS+ etc.
My own feeling but we've no evidence for this currently is that 200ms for cloud to cloud cluster is low, with no last-mile thin-pipe to deal with, over time I want to have our QA validate cluster perf over higher RTT data-paths, buts that's not something we have to time to do today.
In a doc I write 5+ years back, CPPM clustering I did provide guidelines for bandwidth, can I suggest you start their for guidance.
https://support.hpe.com/hpsc/doc/public/display?docId=a00100359en_us