I've said several times that I don't regard measurement-based evaluation of components is entirely "settled science." You can imagine, then, how pleased I was to read an
Audio Application Note on Benchmark* that essentially said the same thing. Measurements are great, but there can be things we don't, can't, or don't know we need to, measure that can have profound consequences on the reproduced sound.
*Some wise and kind soul linked to this article earlier in this thread, but I'm (a) too lazy, (b) too stupid or (c) too full of cabernet to find it and give proper credit.