As I’ve been discussing about this with @Claes also via email, I did the test yesterday and removed two of my four 8GB sticks.
Guess what… Regardless of dt 2.6.2 or git-master, the test acc to this below thread was 3.3% (0.5 s) slower with 32GB compared to 16GB when it comes to CPU only (opencl no effected). This can only be explained with overhead for the mem management.
At the moment my ram is 2133MHz and hence the bottleneck to the 2666MHz spec of my Core i9. As ram prices dropped, so I’ll get my 32GB 3666MHz next week. Will I be too lazy to repeat that test? Don’t know yet 