In a post earlier this year i did provide a paper written and published by Olive that demonstrated the concept that "all" listeners fall into a category of preference for a flat response curve, but also that there are small variations within the range (as to be expected for confidence intervals in any dataset) where preferences included small bumps in treble and bass off the flat response. When you look at the frequency response of really good speakers, those subtle differences in bumps or more humps/valleys in certain frequency ranges and to a certain extent (not too much or too little) can set one's personal preferences specifically to a speaker brand/model OR through the use of sound engineering like Dirac or Audyssey.
In short, i have no doubt that a best sounding speaker is one that is at it's basis 'neutral' and generally flat, but that with small tweaks, people find what they feel sounds best to them (at least to those who aren't trying solely to shop with eyes and wallets as the metric for quality).
