Why is the position of the overexposure limit different for the layer histograms?
Histograms of lighter images are shifted to the left by their EV difference to the darkest image. Ideally this should bring all the histograms to coincidence (of course the lighter images are partly outblown). This provides a visual feedback of the exposure detection process and makes the Separation blending method with its separation sliders more intuitive. With RAW images this works very good as their data is linear (or nearly linear) by nature – the histograms have an identical shape. With JPEGs you will encounter different results, i.e. the histograms will not line up as nicely. The reason is that JPEG data is not linear.
