I don't think it's slower. The cards are designed for dds masked things. As long as there are not too many overlapping alpha levels, it can be rendered fast. With masked dds textures, there is no alpha but the alpha channel will be used only to interpolate the mask contour.
For Shadow volumes you may use a system that allows to A use the orginal mesh for the volume creation, or B use a simplified hidden clone that fits the stencil shadow requirements.
For Shadow volumes you may use a system that allows to A use the orginal mesh for the volume creation, or B use a simplified hidden clone that fits the stencil shadow requirements.