Thank you for your continued patience and those that sent sample traces to aid investigation. We looked into this issue more closely last week and were able to identify a configuration change that inadvertently reduced the per-node throttle in place. As called out, this would have impacted consumers that were relying on persistent connections or keep-alive more heavily.
The problem has been corrected so quota limits should be restored to the previous behavior. Please reach out if you continue to see issues.