Is there a speed difference when using vertex normals instead of face normals?
when using face normal (ie smooth mode == GL_FLAT) the lighting stage have to compute only one color per triangle (3 for GL_SMOOTH).
obviously, when using face normal, there is just one call to glNormal (or one access to the normal buffer when using arrays).
so it is faster to use face normal but since lighting is now done by the gpu (hardware T&L), I think there is no speed difference between the 2 modes.
but if you use a software renderer (or a pre-geforce hardware), using face normal and GL_FLAT smoothing mode is faster.