I guess you are using some sort of API to time the delays? I have seen no noticeable difference (by eye) between 2015, 16 and 17 when adding dimensions, but I guess half a second would be tricky to spot easily. I think I can see a slight difference between the red cross vs green tick scenario though.
If you have stats off the back of API, or some other means of timekeeping, then I think this is worth submitting to your VAR support. They can then escalate to the Developers who write the code, as if they are not made aware then it can never be solved.