Since the question of how to perform a proper test was never answered directly, I'll give it a shot. I started thinking about this months ago and started a thread about it... lots of discussions, few good suggestions (IMHO). Having thought about it more, I think perhaps the problem isn't so insurmountable. And in a very practical way, as per my post to sanjay above, perhaps the application goes beyond testing to how we might approach an actual listening space.
Absorbed sidewall reflections vs preserved sidewall reflections... To me, the easiest way to tackle this is to replace a physical reflection with a simulated one, by placing speakers at the first reflection points in a room with very well damped sidewalls. The actual speaker type used might have some effect on the outcome, but I wouldn't expect very much, so long as the wides are identical to the fronts.
Creating the simulated reflection shouldn't be difficult. There are off-the-shelf processors that allow you to add variable delay to a signal. Dabbling in the recording/mixing side of things as you do, perhaps you already have something with that capability. If not, a minDSP could probably handle all of this easily. A copied, slightly attenuated, appropriately delayed version of the respective L or R front channel is a good first approximation. This should be quite easy to test by muting and unmuting the wide channels. I have no doubt that even blinded you will know which is which, so I don't know that this control would remove your existing bias, even if you tried to randomly select on or off wides. i.e., I think in your specific case, there is likely no use for blinded listening. But you could certainly do this for friends and others that may not have recently participated in such a discussion, may not have thought nearly as much about the topic, and indeed may not have the existing biases. Ideally, you may even want to tell them you want help testing something else, like an EQ tweak you have made, or two different surround processing modes being applied to only the front channels (i.e., tell them the wides are off the whole time). The best controls are sometimes when the subjects being tested aren't even aware of what they are being tested for, especially when audibility is established and it is preference we are after.
This testing could be expanded by modifying the wide channel signals. Change the EQ to simulate front speakers of differing off-axis response. Mix in a copy of the contralateral and or center channels which is more attenuated and delayed to move beyond a first order approximation. Do we like reflection of only the ipsilateral side (ala ambiphonics) or like the increased richness (or muddiness) from the crosstalk? How does our preference shift with levels of delay or gain of the reflection? Are preferences the same for "stereo" vs "multichannel" playback? Lots of things could be tested this way more thoroughly than I find in the current literature.
Cost should be rather low. Room treatments to sufficiently damp natural sidewall reflections (I have no space I can currently do this in or I would consider doing it myself). Another pair of identical speakers and amplification. Processor, whether off-the-shelf, PC, or miniDSP. Might be doable in a budget fashion for a grand or so. Perhaps a few grand to ensure quality speakers and adequate treatments. Certainly not a fortune.
The best part, as per the above post to sanjay, is that this reasoning might be extendable to real listening environments as processors improve. And that means more treatments to create the dead environment needed.... equals more business for you!
Hey, maybe you should champion this approach and hook up with DSX or a studio processor solution to help sell more acoustic treatments. Edited by Bigus - 10/5/12 at 10:26pm