Some time ago I was intrigued by the result of oscillating dt over the maximum height of jumps, like some users have reported. I had the idea of trying to smooth dt over time to reduce oscillations, and googled to see if there was something about this on internet. I ended up finding this very interesting read:
http://bitsquid.blogspot.com.br/2010/10/time-step-smoothing.html
It's quite easy to implement and maybe can have a positive effect on the visual "jerkiness" of oscillating fps. I made a test with a much more rudimentary smoothing than the one described in the article, and at least for the varying max jump height problem it improved considerably (note how the orange dt max height oscillated a lot, while the blue smooth dt max height stays almost constant).
Maybe it could be implemented in C2 with an option that goes from 0 to 100 (that lerps from the original dt to a smoothed one) allowing to choose between compromising distance versus timing precision.
I also had stumbled across these articles describing a method of how to make a fixed time step properly work in different refresh rates:
http://gafferongames.com/game-physics/fix-your-timestep/ (see "Free the physics")
http://www.koonsolo.com/news/dewitters-gameloop/ (see "Constant Game Speed independent of Variable FPS")
They basically talk about "decoupling" game logic updates from rendering, so that the game step can be constant while the FPS can vary freely. I think it could help with the problem of missed collisions on different hardware as well. But this method would probably require deeper changes to C2 that may not be that easy to change.