Like me, you probably wouldn't give a lot of thought to a 7x50 porro since a 7x42 roof is more portable and gives a 6mm exit pupil that's going to be adequate under almost all conditions. I have an old Nikon 7x50 marine porro that's got very high quality optics but is only single-coated (not sure the prisms are coated at all). Consequently it's obviously dimmer and less contrasty when compared side by side to a modern instrument with full multi-coatings.
However, something interesting happens when I compare the 7x50 porro to other binoculars when the day fades into deep twilight. Even though the image is definitely dimmer, less contrasty and narrower, I consistently see more detail under these conditions with the old 7x50 than I can with a Zeiss 7x42 Classic or a Minox 10x52.
I've tried the 7x50's against a range of high quality binoculars now and my conclusion is this; as it gets dark your eyes and brain have to work hard to interpret what you're looking at. A large porro design like the 7x50 gives the brain more information to work with because the objectives are widely spaced and give a 3-D effect to the view. This makes it easier to separate low contrast objects from one another in very low light. Roof prism binoculars tend to have the objectives close together so that 3-D information isn't available to the brain, leaving you relying on contrast differences alone to separate objects.
I may of course be completely wrong about this (it can happen!) but it certainly makes me re-evaluate the usefulness of a full-size 7x50.
Now, I'm not saying a 7x50 is a 'better' binocular than others and I'm not knocking roofs (I own several). I just thought it was an interesting observation.
Anyone noticed a similar effect?
Binoculars certainly all have a point where it gets useless when nearly dark, and that point is different for every type of binocular. But brightness is usually the main factor. As you have found, other factors also are at play.
There is a simple way to test your idea about objective spacing. Just repeat your observations using only one eye to remove the variable of stereo information entirely.