I saw your post in the other thread and tried using your docker settings to get there, but wasn't able to get mine higher. I just took a look at that system to verify no power management stuff was on in the BIOS. It wasn't, but I noticed the DRAM was running at 1333 Mhz. I'm going to check and see if that might be the issue later tonight.
sometimes it's fiddly. try adding or removing a thread to one cpu or the other. sometimes that helps, randomly.
also, if one docker instance is running slower than expected, restart it, and see if that helps.
In testing, I tried one docker instance per thread (real pain in the rear). first cpu, half the threads went super slow (20% below normal). second cpu, all threads ok.
I sorted the threads by speed, and restarted half of the slow ones. Checked speeds, all the threads on cpu1 were now running fast, even the previously slow ones I hadn't restarted.
Bottom line, the whole thing is fiddly. Sometimes all thats needed is to restart the miners.
The majority of the ram I own is 1333, so most likely most or all my miners are using 1333. So I doubt that's the problem. I did notice with monero mining, my servers with 8gb ram were more likely to need me to restart xmr-stak multiple times to achieve "normal" hash rates, compared to servers with 64gb ram. Maybe with low ram, the process is less likely to successfully pin the ram to the correct numa node? At least for monero, restarting the miner a few times (and checking if performance meets expectations after the restart) eventually fixes it.
It's all fiddly, so fiddle with it a bit : )