Looked again, this time with a little more sleep. The bars are getting spread a bit more underwater, which means the topside half of the pic is a bit wider than the underwater half. So some angle is being lost to the diopter. My bad conclusion this time as I'd been up very late hunting for the old Kodachrome slides, doing the scans, and finally posting them. But the shots at least truely confirm what is happening as all other things on both halves of the shot are constant.
On the positive side, it will keep me shooting diopterless with my 17-35's. And using manual focus where ever possible.
Thanks for calling my attention to my bad analysis this time. Its how we learn. In all the hunting for the slides, I actually found a split diopter but hadn't had the time to put it on a lens, which I should have done.
Some of us open our mouths only to change feet! Mea culpa.
Did you actually measure the difference with and without the diopters on your 17~35mm lens on land? I tried this at 50cm distance (image plane to subject) in order to reduce variables and because 50cm is a not unreasonable closer distance to work at (its in the same distance ballpark as a large dome's virtual image of an infinity subject). My findings were based on actual measurements to the nearest mm of a ruler placed at 50cm from, and perpendicular to, the image plane in the camera - I could actually post these but they'd be pretty boring and I see little point in doing so in all honesty. The percentage I quoted is quite simply the ratio of the mm visible with and without the diopter presented as a % of the view without the diopter - very straightforward.
You are right that in that as noted in Ray's book, in theory using a diopter will actually result in an increased angle of view so the lens should appear wider. However, the reason that this does not happen is apparently because the wide-angle lenses we use, are in reality complex designs and the aperture stop is buried deep inside them - it is apparently this which is the limiting factor to the field of view using a diopter and it limits the angle of view because of the physical effect that it has upon the lens's ability to view the image projected into it by the diopter (at least that is how I interpreted my friend's explanation, which he was able to state immediately as he appeared to view it as a straightforward lens designer's problem!). I studied optics a long time ago as part of my photography course and I'm a bit rusty to put it mildly. I tend to accept my friend's explanation - he is after all a lens designer and deals with optical theory every day. I should also add that my tests were all carried out with L series fixed focal length lenses which are more complex and physically larger than many slower designs. Apparently the field of view would increase if a wide-angle lens with an aperture stop close to its front was used - but such lenses apparently don't exist for dSLRs. It would though be interesting to try slower lenses as they may just exhibit less loss of angle of view.
If you look at B&W's (Schneider's) website and look up their close up lenses you will find that their recommendations are generally that these are for use on 80mm focal length lenses and longer, although they do accept 50mm for more powerful close-up lenses and even 35mm for the +5. If there were no problems with using close-up lenses (diopters), why would they make such recommendations?
I do have to agree with Stephen's analysis of your posted images - which basically means that for any really objective test, the scenarios need to be worked out and controlled very carefully indeed. I'm just amazed that such an obvious optical problem has not been fully explored before, but then why would it be - its hardly a test that would occur until the desire to see the differences between various very good wide zoom lenses could be carried out and in all honesty that is only just starting to become important now that there is a choice in DX and FX high megapixel cameras.