The problem is, a single run of the experiment is inconclusive because there are so many variables, including the weight of the knee presser, the body position of the prone guy, the exact placement of the neck, the level of stress at the time (a heart pounding under the tension of an arrest might be more vulnerable than in a courtroom), any personal health conditions not known at the time and undiscoverable through autopsy (doctors on other sites have suggested a number of them, but I didn’t save that link), so on.
You are making my point for me. The question is whether a reasonable man would expect such a tactic to result in a death, and it's looking like the answer is going to be, barring extenuating circumstances of which a man could not be expected to know, "no."
A reasonable man would not have thought that such actions would result in a death for a normal ordinary man. No mens rea, no murder.