I understand your problem, but it's not an accident that PalApply works this way. The output is like that to match how alpha masks work in OpenBOR, which can reproduce the original image's partial transparency by combining the regular output of PalApply with the alpha mask that PalApply...