I have no idea what is happening there... I'm stunned, ashtonished
edited______
As I understand, there is no difference in the static table of the effect because of the shader model version but its size. The rendering is an instant action, each draw call draws the render target inmediately, there is not relation between frames. I can't understand how can be possible that the lights have a different position relative to the shader model version.
The only thing I can think on is that you are using other surface position but the world position, such the incoming vertex position in object space. But it should fail in both shader model versions...