Most of those focus tests are designed so that you're focusing from scratch, versus a focus distance that's close to the last time the camera locked.
Even if a camera loses focus during tracking, very little movement of the optical groups for focusing may be required for a new lock , since the focus distance hasn't changed much from the last lock (unless it tries to lock on the background instead). lol
That should be much faster compared to the way they would probably test them (making sure the lens is not focused anywhere near the correct focus distance when they start timing it).
Once you get an initial lock (unless the camera decides it can't find the target and needs to hunt over a wider range), the mechanical aspect of it would probably play a smaller role, and the speed of the algorithms would probably play a larger role.
That's one theory anyway. lol I'm sure the engineers know for sure. That's one reason I couldn't help but wonder if the new user definable focus parameters in models like the EOS-1D Mark III could be playing a larger role than people may think (processing speed may be more critical than the speed of the focus motors, etc., once a lock is acquired, since not much movement should be required to maintain focus).