When I run coverage over Python's stdlib (this is directly, not using allcoverage since I forgot to use the script), I end up w/ nearly 34 GB of disk space used. Now I really doubt that the .coverage file is using that much space as it was not that big at the end of the run.
So my hunch is that there is a fundamental memory leak that is leading to a massive amount of swap being put on disk. This happens on both OS X Lion and OS X Snow Leopard. When I run the coverage again using the pure Python code then it never seems to trigger a use of disk (other than writing to .coverage).