One of the effects (rolling resistance) is indeed well understood and easy to model. The other one (estimating the frontal area variation vs body weight and height) cannot be done other than as a gross average approximation. At high speeds, the latter is larger than the former.
Yeah, this seems like something pretty simple to test in methods applied by Eric Schlange of ZwiftInsider. Set fixed power (via their special methods ), then bike and rider height as constants. Alter weight in steps on a repeatable course like Tick Tock, which is largely flat. Record lap times and compare to see what if any affect Z applies to weight with all else being held constant.