Because I have much more time than sense, I've done some more work on the script that counts/collects the characters used in files.
Building on the core that Man Eating Duck posted, this script will work for a single ePub, (x)html, or text file. In addition to filtering all of the html code/attributes from the results, it will also convert entities (named or otherwise) to their rendered equivalents.
It also has the ability to limit the results to a single specified CSS class (handy for determining the font-subset required for headings or drop-caps).
Python will almost always have issues printing certain unicode characters to the console on Windows OSs, so Windows users should consider just writing the results to a file and then viewing that file with an editor that supports the required character encoding.
Should work with Python 2.5 - 2.7 (maybe even earlier).
A test xhtml file is included for testing/benchmarking purposes.