For us, we needed to give our scene graph objects a coordinate frame where the rotation, scale, and translation transformation components are kept separate from one another; this is something similar to what is described in the book 3D Game Engine Architecture (http://www.amazon.com/Game-Engine-Architecture-Engineering-Applications/dp/012229064X).
XNA didn't quite fulfill this niche, which is understandable as I imagine one constraint during its design was to make it easier to use.
So, in the end, I'm thinking about writing my own 3x3 matrix struct, and my own Transform struct (which represents the above mentioned coordinate frame construct). Has anyone run into any problems integrating their owns numerics with XNA's