This time I compare the same dog model in its original MS3D format against its MD3 version. The MD3 format is the old vertex-based Quake3 format, while MS3D is the native format used by the MilkShape3D modeller. The latter uses skeletal animation.
MS3D above, MD3 below
In the screenshots above, you can see two scenes rendered at the same frame rate. It’s evident that the engine performs the MS3D rendering at poor speed, compared to MD3 rendering, in fact the first loses 25 dogs against 225!
The main reasons are two: MS3D, being skeleton-based, require more computations than MD3 to place each vertex; then MS3D does not support yet the clone function and every dog in the picture is a different model with its own animation data, while similar MD3 models share the same data, so the impact on memory allocation is reduced.
I think there is room for a lot of speed improvements in the MS3D animator, but in the meantime I suggest, if you own some MS3D model, to convert them to MD3 (if there are only a few animation sequences) or MDL, because they are currently faster.