In John Atkinson's January article about the "Cocktail Party Effect," he references a Wikipedia article that claims, "hearing reaches a noise suppression from 9 to 15 dB." Wikipedia also references an MIT Media Lab paper: http://www.media.mit.edu/speech/papers/1992/arons_AVIOSJ92_cocktail_party_effect.pdf . In that paper we learn that "Under ideal conditions, the detection threshold for binaural listening will exceed monaural listening by 25 dB."
The human brain is a very sophisticated signal processor that can learn to perceive, among other things, very small levels of distortion. A couple of years ago I was initially pleased after acquiring a stero pair of the latest model of speakers from a well-known high end manufacturer. But after a few months, I realized that the speakers presented unacceptable (for me) levels of congestion on some orchestral passages. It took time for my brain (which is, by definition, a learning neural net) to create the signal processing to perceive this shortcoming.
Consider a a particular recording and a particular set of reproduction equipment that just happens, simply for example, to reach audibility of distortion at the 5% level upon initial hearing in monaural for one specific listener. Perhaps after stereo experience has trained the ear-brain of that listener (whether this training takes minutes or months), he or she can achieve something like the 25 dB sensitivity improvement for stereo detection threshhold sensitivity to distortion (I understand that I am not applying this number in the way it was measured, this is speculation for the sake of forming a hypotheis that may be either true or false). 25 dB below 5% voltage distortion is now about 0.25% distortion, if my mental arithmentic is working properly this early in the morning. For those who can perceive 1% monaural initial distortion, this might drop to 0.05% by the same arithmetic.
Some possible hypotheses suggest themselves to me: One is that our natural neural nets can learn over time to perceive things that we do not initially hear (for example, within the time frame of a Double Blind Test as it is usually administered with equipment unfamiliar to the listeners). Another is that we may learn to reach levels of discrimination in system performance that have never been suspected by those who make traditional DBT measurements. A third is that the process itself of performing a DBT may intrude upon the implementation of our mental signal processing at the highest level of performance. These are only hypotheses, but they suggest to me that a better protocol is needed than DBT as we now often use it if anyone is serious about the science of measuring audibility of equipment effects. From experience, it is my claim that a human brain can become trained to perceive equipment differences well below the levels that have been demonstrated using the low sensitivity method we call DBT. I have no desire to fan the fires of believeing that every tweak of any kind makes a big difference in audio system performance. I do have an interest in understanding more about audibility, and I wish the experimentalists in that field good luck with their very difficult challenge!