Hi Steve,
As Jeff quickly noted, the notion of "measures well" will be a different set of metrics and measurements to each designer and enthusiast. Some definitions will run quite parallel but differ on prioritization, weighting of one metric over another, or have a few different points of interest in finer details, while others will seem to be diametrically opposed. The inclusion of the qualifier "always" in the posed question effectively insures an answer of "NO!" Absolutes are rare in audio. Good guidelines and indicators are much easier to pin down without having to list a hundred exceptions.
IMO there are two primary problems which hinder further rational discussion without much more narrowing and specificity of what is being discussed. First is that most all loudspeakers defy description with a single dimension metric. There are many more dimensions of variance as compared to electronics making for a large matrix of variables. Even if we pick a set of measurements, determining how to prioritize and weight them, let alone defining what's "good enough" makes for many variables to consolidate. Secondly we have to acknowledge the degree of imperfection in real loudspeakers. It is much easier to define a flat line magnitude response into a resistor as a target for an amplifier than the magnitude response of a loudspeaker at some distance and angle. With such an array of variables, and significant imperfections or variances in most of those variables, you could easily devise a metric which results in the same deviation but dramatically different results.
Here is one conceptualization I like to use for acoustic measurements of loudspeakers when fielding such questions is to imagine the range of audible frequencies as the spectrum of light. I have come to realize I'm unlikely to have time to spend time on this myself anytime soon, so maybe Dr. Olive or others would have time to do some tinkering with this simple computer mapping (I'd love to see it)...
- Start by scaling/mapping the visible spectrum of light to the 20-20kHz range of frequencies.
- Set the intensity range to vary from max to min over a +/-10 to 20dB window (20-40dB).
- Generate a color based on the exact frequency response input to the system as a mix of the light spectrum. (ie a subwoofer only would result in red)
- Map a grid, circle or 3D sphere to a set of on and off axis measurements for visual display and interpretation.
Common spectrograms for off axis behavior do part of this for a single axis but instead display color as intensity vs. the spectral deviation I expect the above to display. The above would consolidate the horizontal display of a spectrogram into a single color. In simple terms, it gives us a familiar metric, color, by which to differentiate the 100s of different curves we can draw within a +/-2dB window. At bare minimum, I believe this representation might better convey the basic concept that all responses which meet +/- however many dB are not all equal.