The current ray-object-intersection algorithm is pretty poor, because the algorithm computes and yields a list of ever object which is intersected by the ray. Not only that, but that list is also unsorted, meaning it has to be manually sorted based on the raycast result's time of impact, in order to determine the closest hit object.

A new algorithm should only do the bare minimum tesing for retrieving the closest / first object hit by the ray.

This issue would be more or less fixed by the ncollide issue #88, ie. awaiting best_first_search to be implemented for DBVT and exposed in CollisionWorld.