The speed improvement is great, but pmdesc2 has still some flaws:
(1) Version numbers with more than one separating point (sub-subversion) will be converted into control characters.
(2) DOS-terminated files are not handled correctly (no description found).
(3) The extracted description is not limited to a maximal length, so faulty docs are slurped in completely.
(4) Some modules are split into a pm-file and an accompanying pod-file which leads to double entries.
I rewrote parts of the script and called it pmdesc3. It overcomes the above mentioned flaws. Output can be sorted and formated. It has a POD included. pmdesc3 is part of the VIM-plugin perl-support.vim (script # 556) and can be found at http://vim.sourceforge.net/scripts/script.php?script_id=556