The way you're calculating the expected direction could be off. Also increasing the simulation iterations could improve it.
I've gotten a close before, but not completely exact which I attributed to the collision time being approximate.
Unfortunately the capx is very wip and is left in incomplete state. Also it's a chunk of math to calculate the collision time and the result of a collision which I am no longer familiar with.
https://www.dropbox.com/s/yz18begzbo294 ... .capx?dl=0
Also here is some reading on the math it was based on:
http://www.real-world-physics-problems. ... iards.html
I just realized the capx uses the built in physics and not chipmunk if that matters. It's from an old attempt at the problem and I probably won't have the time to refine it. Basically it would be perfect if we replicated the physics simulation and advanced the simulation to see the path.