Hold a piece of reasonably long (2 meters) drywall, which typically has a paper surface that provides an excellent tracking surface, between the two mice you wish to experiment on, and let it slide free while using Mousetester to record the data.
You need to keep minimal and consistent pressure/angle throughout the fall to ensure the acceleration is only attenuated by a systematic amount. The heavier the drywall the less error is introduced by friction.
Mathematically, it should be a piece of cake to infer the velocity throughout from elapsed time and modified net acceleration. I actually recommend curve-fitting with an acceleration parameter so that you can make multiple trials and normalize them. This will pretty much be equivalent to turntable tests given enough trials.
Responsiveness (slightly uncertain human error margin though):
Same setup, but this time you are free to modulate the velocity via changing pressure or angle. It is always easier to tell phase differences when the motion are modulated.
As an aside, I propose a slight change to the current methodology of responsiveness test: instead of a single swipe, execute a periodic, rectilinear motion, and average the delay at the peaks. Higher frequencies are easier to analyze -- within reasonable bound in terms of speed and distance of course.
It is a truth universally acknowledged, that the comfort of a Logitech shape, is inversely proportional to the R&D put into its making.