Wow. Using zram and putting swap into ram. Completely silly.
So now you swap more often, as more of your ram is gone, but you're swapping to the very ram you shrunk.
Not to mention, ram access is indeed limited by ram speed, cpu cycles, and more. Slowing ram access down by taking kernel caches, and forcing them to swap, via a virtual drive, means you're slowing down access to main ram, while a swap of death happens in your ram disk.
That's not how it (ZRAM) works. You're holding it wrong.
Though I have to concede that there is much (mostly) wrong documentation about it on the net. On low RAM systems, where such stuff could happen, it may be insufficient to just use it, there is more setup necessary, beginning with https://docs.kernel.org/admin-guide/sysctl/vm.html?highlight... and associated sysctls.
Also tools making use of 'PSI - pressure stall information' are lacking, especially making use of these in an automated and universal way.
So now you swap more often, as more of your ram is gone, but you're swapping to the very ram you shrunk.
Not to mention, ram access is indeed limited by ram speed, cpu cycles, and more. Slowing ram access down by taking kernel caches, and forcing them to swap, via a virtual drive, means you're slowing down access to main ram, while a swap of death happens in your ram disk.