Hi,
I’d like to thank you first off for a great post, along side supporting information. We really appreciate well thought out posts like this!
Group chat is not yet being processed on AWS. Some components of our chat infrastructure are located on AWS now (as you have mentioned). Additional round trips per request increase the latency to send a message, connect, etc. If this was the cause of these specific spikes you have noticed, we have a plan to reduce them 
Do you have additional information about the latency you have recorded? Do they happen periodically (every 30 seconds for example), or do certain actions always take longer the first time? Some increased delay might be due to our caching infrastructure not having a value, or whether they require a write vs. a read.
Thanks,
Xangold