Well done, EF, for putting all this together and getting round to some serious testing! If memory serves, your findings on the flat are comparable with other comparison of recumbents and uprights. As you say, uphill introduces more factors and biomechanical ones are perhaps the most difficult to measure. Unlike Kim, I am sure I still go faster uphill on an upright even though I have several years of riding mostly recumbents. Problem is, it is only subjective, but on a hill that I ride very frequently (roughly 8 - 10% gradient) my speed on the upright is consistently 1.5 - 2kph (10%) faster at a pace that I would describe as 'hard sustainable'. That's the difficulty, though, and you raise exactly the same questions: can I actually sustain more power on the upright because of the riding position? And, if riding a long distance (eg long audaxes) would that greater power output mean that I was tired more quickly because, overall, my body was working harder, or would I just do the same ride quicker (depending on how hilly it was) on an upright. (An academic question for me because my back would soon stop me riding on an upright.)
Assuming the more power-on-an-upright theory is right, there is at least one other possible explanation which is to do with what happens on the two types of bike when you press hard on the pedals. On an upright, some of the energy you produce that doesn't go into turning the pedals might reduce your weight on the saddle; on a recumbent, you just compress whatever seat pad you have and possibly cause slight deflections in the portions of the seat you are pressing against. I suspect that energy spent doing the former is more recoverable (during the next part of the crank revolution) whereas on the recumbent you just heat up your seat. You might be able to reduce this by for example pulling on your arms (depending on what steering configuration your bent has, but I doubt if you can get all of it back.
I can't think of a way of testing this directly but an extension of your experiment might be interesting: you could get several people to do a maximum effort test up a hill (wouldn't actually have to be up a hill but might make it easier to get back to the start) and compare the power they can sustain (eg for a minute or maybe 2 minutes) on a recumbent with what they can do on an upright. You would probably get differences depending on whether they were practised bent riders but if each individual did a few of these all-out-effort tests and you had a few volunteers it might produce interesting results. I am sure others have done similar tests but I have never seen very convincing results, eg very small samples, etc.
Happy researching!
Alf
Let me know if I can help in any further testing - sounds great!