There are three reasons, none of them terribly good in isolation.
- I build business applications, not graphics applications, so the "basic" GTK was precisely what I needed.
- PyCairo is (to me) a fairly recent innovation, and my GTK knowledge is a few years old.
- Cairo requires a more complex technology stack; GTK 2.10 is already complex enough.
OLPC's Sugar specification does include Cairo, so there isn't a good reason to avoid it.
Your point about Cairo is interesting, since it provides a more powerful drawing environment which does very clever things very simply. Some of the drawing power, however, is a distraction from my real purpose, which is to teach programming.
Essentially, programming is all about data structures. In modern languages (Python, Java, C, C++) the "language" part of the education is quick. The data structure part of the language, and the add-on modules are the interesting and challenging bits. This is were the real learning occurs, since the statements are used to manipulate the data structures. For the consequences of this philosophy, see my two boring, useless "Building Skills" books .
This means that I (as consultant/educator) have to dig deeply into Cairo to ferret out the data structure features to show how a Path is similar to a Python list. It has the advantage of making the list structure immediately applicable to a Path.
Additionally, there seems to be a pleasant subset of Cairo features that mirror the basic drawing features of a more naked GTK.