Will mediump and lowp precision have any positive performance impact?

GLSL’s mediump and lowp precision qualifiers sometimes offer a performance uplift on (some) mobile GPUs but (as far as I remember) nVidia’s GPUs ignored it or they don’t have support for something less than FP32. Is that still the case? Does it make sense to use (when applicable) mediump/lowp for nVidia or it’s a waste of time?