No, I was thinking about the actual (total) rotation angles, like yaw, roll and pitch. For instance, each time I want to rotate an object around the three axis, I set the identity to zero and rotate the object to the value stored on the related hashtable.
It's just that on limited hardware, doing sins and cousines
to get the actual angles might drop the framerate, but now that I think, recalculating the rotation matrix may also do the same thing. (It actually depends on jPCT's implementation, CPU / GPU routines and the graphics driver).