First, we shouldn't confuse Coverity's numerical measurements with actual code quality, which is a much more nuanced property.
Yeah, but good quality might well correspond to some sort of measurable anyway. Provided you've got the right measure. Maybe some sort of measure of the degree of interconnectedness of the code? The more things are isolated from each other, across lots of levels (in a fractal dimension sense, perhaps) the better things are likely to be.
Maybe that would only apply to a larger project, and I'm not sure what effect system libraries (and other externals) would have. Yet the fact that it might be a scale-invariant approach makes me a bit more hopeful, as it wouldn't be so susceptible to the "ravioli code" problem, where the code's nicely packaged up into little pieces, but the pieces interconnect in a horrible mess of higher-level spaghetti code. Worked on a large project? You'll have probably seen it in the wild. (Yeah, I've had people argue to me that their code didn't use goto and so it had no spaghetti code problems, despite the fact that everything was so nastily interconnected that nobody else could understand it. If that's not indicative of a problem, what is?)