If efficiently sorted, I see structures that tend to be 4-7 labels broad, 1-3 layers deep. It makes sense to subdivide emails into relatively similar group sizes, e.g. if you have 1000 emails and sort by N labels, the number of emails per label should be roughly 1000 / N.
However, certain categories are more email-intensive than not, e.g. your PhD students and projects, or work with your foundation probably generates more email than distant family relatives per unit time.
The other metric to optimize through labels would be "ease of search", or to be even more specific, minimize the number of clicks and required text entries in search to find a given email.
If you use Gmail client (which I think your blog post mentions you do), you can name labels "1 - labelname", "2 - labelname" to force ordering. Otherwise, the labels are ordered by alphabetical order, which may be suboptimal.
I'd share my label and sub-label titles, but a student or clinician's labels would be much different from those a CS faculty member would use :)+James Malcolm
surely has some pro system.