You can't get less DR, because you are not changing the ISO. You are
recording data which normally gets clipped - if you can turn it into
useful detail, that's more DR. If you discard that extra data, that's
exactly the same DR as with this feature disabled.
Compare the shots at the same ISO, shutter and aperture (only change highlight++ setting). The midtones and shadows are only darkened (to make space for extra highlight data), so you will have to brighten the image in post. Brightening will NOT introduce more noise than in the original shot (with highlight++ disabled) - and THAT's the point. If you would underexpose in order to get more highlight detail, you would get more noise in shadows.
So, for testing:
- Make sure you have fully manual settings. No ALO, HTP, peripheral lens correction, auto WB or other things like that.
- Take a picture which is slightly overexposed (for example, set zebras at 99%, increase exposure until they start to appear, then overexpose by 1 stop)
- Record this image with highlight++ disabled, then at 1EV, then at 2EV. Don't change other settings.
- Brighten the darker images (use the Levels tool to clip R, G and B channels at the same point - just before green starts clipping) -> with this you should get the original image (well... almost, since the response curve is not quite linear; a flat picture style may help).
- See how much detail can be recovered.
Now I'm thinking to reverse engineer the response curve and get the RGB data back to linear if possible. This would make postprocessing a lot easier (or at least the research part).
Here's a small howto for obtaining a mask for overexposed highlights in Gimp:
- Decompose the image to RGB (Colors -> Components -> Decompose)
- Look at the histogram on the green channel (or the channel that clips first) and see where it clips (it will be under 255, obviously). That's the white point.
- Threshold R, G and B channels at the level you have just found. Combine the results with "Lighten only". That's a black and white image - black where color data is accurate, white where it's not. Blur it a bit.
- Optional: use Curves (or Levels) to clip R, G and B channels at the white point. You should get roughly how the shot would look without this feature. Recompose them to get a preview, and compare it to the overexpoesed original shot (taken without highlight++). Discard this image, it's just for your info.
Now, you need to guess the color in the areas where the image is overexposed. Use your skills for this :)
And, finally, blend the two layers (the original one and the one with guessed colors) using the mask. I have attached the result of blind desaturation in the overexposed area => where the subject was already white, it's fine; where it wasn't... it's not that fine :)