One of the defining characteristics of fractals is that they are self similar, meaning that a small part is similar to a larger part, or to the whole. The authors argument is that there is a self similarity in the quality of restaurants, by looking at a small part (in this case the napkin) you can with a some certainty say what a larger part is like. Fractals don't have to be geometric patterns, in this case it's the quality that's self-similar. Fractals aren't pictures, they're a mathematical concept. In this case I think the analogy is a good one.
They're also non-Euclidean geometric shapes. So calling it fractal is ambiguous at best and requires you extract one element that makes a fractal. While there are times writing in such a way can be helpful, this is not one of those times. Instead, "Quality is consistent across levels" or even "Organizational quality is self-similar".
I didn't think fractals had to be geometric shapes, but Wikipedia seems to agree with you, so I stand corrected. (http://en.wikipedia.org/wiki/Fractal)
Uh as you admit below, your argument here is mistaken.
Fractals are geometric patterns only. The restaurants are not "self-similar" in any geometric or fractal sense - the napkins don't particularly resemble the steaks. The only thing a good restaurant has a consistent quality. They don't resemble fractals any more than steaks resemble chairs and napkins.
With a reasonably defined quality space, what the article is saying is that the measure in any quality dimension is likely to correlate highly with that in any other: i.e. the measures are self-similar.
