We also have epyc workstations and we found a similar issue but not exactlly.
Actually we are vendors producing clusters. In our cluster (centos) with amd epyc CPUs, we test Fluent 2022 and they scale extremely linear. 10 nodes, 10 times faster. So back to your topic, I dont think its a problem of Fluent.
On the other hand, we are also facing a similar problem: Fluent 22 is slower than 20, but it only happens on windows system.
Do you have any updates for your issue?