Technically, using a pixel/vertex shader can be faster and more efficient, because it customizes the rendering pipeline, overriding the actual FFP processing. For example, int this way you can prevent the device from calculating dynamic lighting (static sunlight only, or fewer than 8 lights in your shader would be a speed increase over basic FFP). This is in general, I'm sure there are some inconsistencies with older cards due to the nature of the first pixelshader architectures paired with slower card processing speeds.

It's really up to you, I'd try both, and see what works best on a wide range of systems...

-Rhuarc