Interesting idea, but likely ineffective for many of the reasons stated. And there's the possibility of negative transfer. Other than reinforcing the idea of the "shopping cart" example, there's not much to be gained from this in my opinion.
Many, many moons ago when I taught in helicopters, a fellow instructor was teaching a student when a hover-autorotation accident occured. This resulted in damage to the helicopter's undercarriage. The flight school owner didn't fly helicopters, but nonetheless designed (on paper) a contraption which used a seat, counterweights, pulleys, and a collective-like lever to "simulate" the mechanical inputs necessary to manage the hover auto. He pitched the idea to the flight instructors, myself included, none of which thought any good could come from it, and the idea was ultimately abandoned.
There's good training, and there's bad training. There's not a lot of "okay" training. Things that fall into the gray area (maybe this tailwheel "trainer" idea) will generally move the needle in the wrong direction in the big picture. Just my opinion after a lifetime of doing this stuff.