If that means that you know something relevant to this that I don't know perhaps you would be kind enough to share it?
What I don't understand is why you are having so much difficulty with the concept of system response time constant, and the widely recommended input sampling interval of 1/10 that time.
Your observation, quoted below, follows directly from a basic understanding of the above. There is no need to experiment and rediscover this basic, very well understood principle!
My practical tests have suggested that it makes no sense to read the temperatures faster than they can change.