Thank you for the additional background information. I now understand better what you are doing, and what your references are.
I hope I'm not overstepping your bounds to make additional suggestions and send additional references.
With respect to the separate transform object, I agree that you wouldn't want to create a new transform class. And, certainly, it does not necessarily make sense for OBB to inherit from transform. The OBB could have a property that is a transform object, and I wonder if it might still make sense to take advantage of the existing scene graph transform class. If you do something like this in your collision detection engine, then it might make it easier to sync your collision detection with the scene graph to be rendered. Or, the OBB shape could be a child of a transform object...making it easier to support compound shapes or different types of collision shapes. Many physics engines are set up sort of like this. I realize that rendering engines and collision detection/physics engines often have separate representations of the scene, but the collision/physics engines usually provide helpful built-in mechanisms to point back to a rendered scene graph.
I can think of another OBB reference, which might be useful if you can find a copy online. It is the OBBTree paper written by researchers at UNC Chapel Hill many years ago. Yes, its really old now, but is one of the early presentations of OBB for collision detection, and still useful. Here is a link to the paper. (I admit that I haven't ready it in a long, long time.)
Pierre Terdiman also has done some work that you might find useful:
http://www.codercorner.com/Coder.htm - see OPCODE for example