This was a massive reencoding effort and the details of what tools were used are available at http://internetarchive.wordpress.com/2008/11/25/fast-and-reliable-way-to-encode-theora-ogg-videos-using-ffmpeg-libtheora-and-liboggz/
A setup to get the job done. However, after having had a closer look I discovered that the Vorbis encodings were done with e.g.
ffmpeg -y -i CapeCodMarsh.avi -vn -acodec vorbis -ac 2 -ab 128k -ar 44100 audio.ogg
Ooops. This means there's massive amount of content, encoded with libavcodec's Vorbis encoder. The only problem: That encoder is rather primitive and produces significantly inferior audio quality compared to libvorbis. And yup, the Archive.org encodings sound rather unpleasant despite not starving on bitrate.
How I wished ffmpeg would loudly complain whenever using unfinished/experimental/not-state-of-the-art encoders.