I think the best next step here is to set up historical monitoring of your system so that we can figure out where the bottleneck is, because we’ve established that it isn’t CPU time. It’s always possible that you’re maxing out your network connection!
plus more traditional server metrics like node-exporter.