This is the end of this expedition. There are a lot of things that I didn't had time to try. Here are a few of them.
As we saw, the number of external symbols is quite big. Even one file system can have hundreads and we end up with more than a thousand symbols when we consider a decent selection of file systems. This makes tracking each symbol individually not very informative. One way to reduce the complexity is to try to classify them. Doing this manually could be accomplished by somebody familiar with that respective kernel but some automatic method might be also attempted. With this classification in place, the way the file systems are using various classes of symbols could become more meaningful.
Another direction, which increases complexity, would be to take in consideration not only that a kernel module is using a certain external symbol but also from how many different places from inside its code it is doing this. This information is contained in the relocation table of the object file and it can be easily extracted using objdump.
From the personal side I can say that figuring how to get all the Linux modules was kind of cool and learning about the way hierarchical clustering works was very informative. The fact that BSDs have a system release with each of their kernels (or conversely, that they make a new kernel release with each system release) made them much easier to deal with. Their archives, which contain binaries going all the way back to the very beginning, represents a very valuable resource which could be used to track their evolution.
Some trivia. There are 78 regular figures (out of which 10 have high-detail versions) and 4 animations. The building of the phylogenetic tree for all the 2.6.x took about tree days of continuous running on a P4 at 2.8 GHz. The memory consumption was decent though, only 200 MB. Before settling to the final Circos graph I generated more than 50 circular plots showing the relations between each file systems with everybody else. All of them look very similar though. Except the circos plot and the the treemap all the others are done in R. The treemap was obtained using GrandPerspective and Inkscape.
Thank you for reading! And once again, if you find any mistake please let me know.