When a light interacts with a surface, Doom 3 engine goes through all triangles of the surface in order to detect which ones are within light volume and front-facing. If the light constantly moves, then this processing has to done every frame. If stencil shadows are used, then engine has to look through triangles once again in order to build shadow volume. This implementation detail means that large static meshes with small dynamic lights nearby can be surprisingly slow. I think this issu