If anyone is interested, I finally ran a bunch of benchmark runs this morning and I have the results here:
Results seem to mostly make sense, though I forgot to include that the client loop time is inclusive of tail latency, even though no strong tail latency was detected during the runs