The misunderstanding here lies in the actually difficulty/cleverness required to create good collision error handling algorithms.
I agree, everyone can track down the variable or whatever that goes crazy in case of a ‘collision freak out’, but then how to handle it? What algorithm to write/apply to not have it freak out? This is where advanced coding/maths kick in.
If this was any easy task that could be done by a single hand, nVidia would have implemented it already. This is the core reason why Havok swims on top. (next to speed, memory and multithreading)
http://www.havok.com/products/physics
Everything written here, especially for collision -> True!