It should have been excluded both times, for sure, and the disparate treatment does probably reflect some AMD favouritism, but if Zen 3 is very close to Comet Lake in GB when you disregard crypto, that just means Zen 3 is relatively weak at GB compared with other workloads. Zen 1 and 2 underperformed in GB too.
I'm not sure it should be completely excluded. AES does affect a lot of the user experience. But in my view, for a client system, AES went beyond being noticeable to a user at skylake. I mean sure, if you open up 20 youtube videos at once maybe it's noticeable. But, who does that?
Just on these benchmarks in general though, there's so much variation from one machine to another I think anything under 5% can easily be due to things other than the CPU.
I mean take the comparison below. These are both 10400's, mine on the left and someone else on the right. Both of these are very high single thread for a 10400, the vast majority of 10400 scores are below 1100. I was curious, how they got more.
Then I noticed that they are running a Unify which is a $300 motherboard vs my Prime Plus which is a $150 motherboard. The memory is almost certainly different. As you can see, the win isn't 100%. I win in image compression, and machine learning, by 7.5% and 16.5% - significant.
Also worth noting, I beat that system in multi-core by 4.7%.
What I'm saying is, you can tune these rigs to do well at different things. Memory bandwidth vs latency, is probably what this reflects. Again, these are the same CPU on Z490 :