Yeah, I can get Physics going in pixels-per-second when I figure out the math, but it would change every existing .cap...
You divide by timedelta because you want to apply a total-force-per-second, not a force-per-frame. For example (picking easy numbers), say you want to apply 10 newtons/second of force. At 1 fps, that's 10 / 1 = 10 newtons per frame. At 2 fps, that's 10 / 2 = 5 newtons per frame, which adds up to 10 newtons per frame again.
Physics itself is a bit fussy about floating point precision and framerates. Generally it's stable in the 30-120fps range. Outside of that I think you'd start getting weird rounding errors in the timestep.