i can't imagine why this would be impossible. judging by the way AR plays, it could be pretty easy. this is hard for me to explain in paragraph form, so here are bullets
some processing power is spent on rendering. cut that out, the physics engine can process each "frame" quicker.
just because it can process quicker doesn't mean the simulation will go faster. it will just go smoother on slow comps. if it AR is run on a fast comp like this, it will run the same speed (fps) but the comp will have more time it's not working.
this means you have to increase the fps proportionally to the veloctiy of everything. for example, if AR usually runs 60 fps and at the beginning dillo is going 5 feet per 1/60th second, you would have to make him go 10 feet per 1/60 second at 120 fps
why double the speed when the fps is doubled? if you didn't, AR would check dillos position at double the frequency, so dillo's position would be checked twice every 1/60th of a second, so dillo would really be checked every 2.5 feet at each frame. that means that if in 1/60 of a second, dillo moved from one side to another of a metal piece the physics wouldn't catch any contact, but if it ran at 1/120th, dillo would be caught and the original physics would be broken. double all velocities and the problem would be fixed.
the |'s are points in time where physics would be done (1 frame)
the X's are armadillos position at that point.
the ^ is the metal bar
|-----| <<< at 1/60 contact isn't caught.
X.....X
. ^
|--|--| at 1/120, contact is caught and physics are changed
X X X
. ^
|--|--| at 1/120th & 2x velocity, contact is not caught (normal) and physics is unchanged
X.....X.....X
. ^