Diomidis Spinellis (author of Code Reading) calculated the following numbers for various kinds of human-generated text in the FreeBSD project:
Asset Source Size (KB) %age
Issues database GNATS 711783 30.5
Source code C/C++ 431519 18.5
Documentation troff, DocBook 114450 5.0
Version history CVS 1076762 46.0
As he points out in this month's IEEE Software article that, this is an indication of just how little of the "stuff" in a real software project is what students think it's going to be.