I plan to measure out a calibration course on a rail-to-trail close to my home as soon as things warm up this spring. I will then do some calibrations on it and compare to calibrations on my usual paved cal course.
But it won't be very conclusive for the general case for many reasons. The two courses are in different locations. One has only a very slight incline while the other has a bit more of an incline. One will probably be a bit more sunny than the other. Calibration rides vary by a count or so from ride to ride. Eliminating this noise from the experiment would require a large number of rides, and this introduces the problem of a changing temperature during the experiment.
Add to all this the fact that there is no one definition of a paved surface and one definition of an unpaved surface. Mike Sandford has shown that calibrations vary even for two different types of paved surfaces. It's almost certain it will vary even more for two different unpaved surfaces.
It would be great if a large number of people did some experiments comparing cal constants for paved and unpaved surfaces. The best setup would be to find a location where paved and unpaved surfaces are side by side, similar to Mike's experiments on different paved surface types. But it's difficult to find that because usually the unpaved "shoulder" of paved paths isn't in very good shape.