Well I will have a chance to test the theory for myself tomorrow. Personally, I came to like the sound of my HF3s better after 50-100 hours or so. (Most of it with me actually listening) Now, whether that was me learning how to use the tips better, and get better seals, etc who knows?
However, I have ordered a second pair, as I wish to leave a pair at work. They arrive tomorrow. So, in theory, I can directly compare a non burned in pair vs one that has a number of hours, and I can interchange the tips to take that out of the equation. Of course, it'll only be myself that I get to convince. And there is always potential placebo affect, however, I'm up for the experiment!
Also, the third part to the experiment (assuming I hear some initial difference) is to AB them at a later date and if they then seem to be the same, that would rule out differences for individual drivers (again not ruling out placebo affect)
At any rate, as a real scientific test of this would it not be possible to run the same piece of music through both sets, and record the output, digitize it normalizing for volume levels and then run a Fourier analysis on the result and compare them, and then run the same tests when the new one is burnt in to compare the differences if any? I don't have the equipment to do this, but I would assume that most high end audio shops would be able to do the analysis and have the ability to do the recording.
Also, the comments on science are amusing. That's having the arrogance to believe we actually know everything. Science is the state of current understanding of how we model the real world. There is always room for error or missed assumptions in the model. (And yes, I am actually a EE so I do have at least a basic understanding of how electrical signals are conducted, and had in fact been much in the anti black voodoo crowd :)) At any rate, the trick is that you always have assumptions in your modeling and they may not be correct. I had someone working with me while we were tracking down a device driver bug in a complex embedded system. His statement was, "it's a computer, it does exactly what you tell it to do" I eventually proved this statement entirely correct. We wrote a piece of test code that over and over again wrote a particular value to a memory location and then immediately read it back and flagged an error if it wasn't the same value we had just written. After not too many iterations the test failed. We captured it on a logic analyzer and noticed a timing violation in the way the write was happening to memory. We were right on the edge of the spec for what the chip would actually handle. So yes, his statement was entirely correct. The system did *exactly* what we told it to do. However, our assumption of what we were telling it to do was incorrect. I have also seen many cases where people came up with solutions / improvements that did in fact work, based entirely on incorrect ideas. However, a side affect of something they did resulted in improvement.