I am not sure if AoTs is relevant for real world game performance at all ? IF there are plenty of threads communicating between each other ( locks guarding updates to same structures from multiple threads, submitting rendering work to DX framework, that in turn deals with multithreaded drivers etc ), you are bound to hit into ZEN architecture limitations due to CCX size.
vs
Kinda obviuos that if the number of threads is > 16, average comm speed on ZEN3 will slow down. It can happen sooner with not optimal scheduling, for example if physical cores are loaded before HT threads that share same L3 cache domain. Consistency on 5950x could be hilariuosly bad too, due to random factors like NV drivers getting scheduled on wrong core complex or esp bad arrangement of critical threads between CCXes. In fact from my MT programming practice, it could be down to just two busy threads, that have bad case of false cache line sharing between them, getting scheduled on different CCXes sometimes and ping pong of cache line owner ship impacting performance some 15% or so.*
The real question is how relevant is this sort of scenario now that AMD is up to a chunk of 8C? Hard to imagine any of current games needing more than 8C of ZEN3.
So in my opinion what AoTS is after this update, is providing insight into hypothetical future limitation that we are already aware from AT testing.
* extra point of fun debugging it when it happens across sockets on MP system.