Yes, I got that, and by now I also see that it should be. I re-read FarbigeWelt's link from the first response:
viewtopic.php?f=2&t=474&start=120#p5737
where I noticed he mentioned it already. But more below...
I also understood that, I was focussing more on the "like" alpha. I'm still not quite clear about how I should understand its effect exactly. It seems only the surface effects are "made transparent", but the volumetric effects remain? But this is not so important at the moment to me anyways.
Now to the "new" stuff (rediscovering what FarbigeWelt already observed in the link above). In short: The glass doesn't work properly unless "rough" (or "architectural" for partial realism) is checked. This is also what is needed to make the blend-file provided by FarbigeWelt in this topic look as expected.
I played around more and tried to make a real experiment which I then recreated with Luxrender. It consists of a glass of water with a few drops of milk as a homogeneous scatterer, observed with a DSLR with 70mm lens from about 500mm distance.
Appended are the following images:
- photograph of the experiment
- Render as best as I could in this time with rough glass (roughness=0.01)
- Render without rough glass
- Photograph and rough-glass render with adjusted histogram to show some contour of the scatter within the milky water.
- blend file
An interesting observation is that I only need to activate roughness on the glass material, but not on the water, which is also set up with a glass material. Maybe that has something to do with how the volumes overlap (as described in tutorials)? Didn't test that further.