You should not really have two full nodes, if your plotter and farmer are on the same local network. Both of those nodes will UPnP compete for port 8444, and part of the problem could be clashes around that.
Saying that, if you really do want to have two full nodes, and want to keep your current setup for a while, I would make your farming node peer with your plotter node only. This will remove all extra peer connections from your farmer, and will give it a 1Gbps connection to your plotter, as such it should never have syncing problems (IMO). At the same time, your plotter will do all the blockchain heavy lifting with those 10-20 connections. I have never done that, but I think that it should work. I think that some other folks may have tried it already.
Also, I would not really bother with manually killing any peers on your plotter, as that can soon become your primary job. Let the protocol sort it out, and just hope for the best.
Your plotter is way more powerful than my full node (low power i5 NUC), so with the number of peers dropped down to 10, the question is what is your blockchain db sitting on? Is that HD, SSD, NVMe? Try to stop your chia, and once down, try to copy that blockchain to another drive. If you get low xfr rate (I would say that everything below 100MB/s is bad on your rig), this is the time to move that db to an NVMe (although, not the one used by your plotter).
I am really not sure, whether those are really nodes that have problems, or rather that your plotter is having problems (this would be my guess), and just cannot chat with those peers for some reason, thus those garbage numbers. This looks like what Larry had (a lot of “bad” peers). In his case all got sorted out, but he also started by killing his old blockchain db, downgrading to v1.2.9, and syncing from scratch. I am leaning to say that potentially that indicates that you may have a corrupt db. Again, I didn’t have that problem, so cannot say that I do know what to do in this case.
The reason that I am saying that the problem with those “bad” nodes rather points to your rig is that in my case, I am fully synced, and I have not one node like that. Virtually, all nodes are synced, or about to be, and just few are really low, it means starting from scratch, as I see my node is sending plenty of data to those. We both have the same P2P protocol, so it is rather unlikely that for whatever reason we would see such big difference in the status of those connected peers.
Although, saying that, I am still on v1.2.6, and Larry did his sync on v1.2.9. That v1.2.11 was a rush job to address Dust Storm, as such maybe it just crippled P2P protocol, and thus are those “bad” nodes.
Maybe folks that are on v1.2.10 and below can chime in, if they see those bad nodes. Maybe this is just v1.2.11 problem.