reference material first:

http://en.wikipedia.org/wiki/Unconference

http://www.unconference.net/unconferencing-how-to-prepare-to-attend-an-unconference/

http://www.unconference.net/

I have attended a number of unconferences over the last 5 years, and find them a richly rewarding alternative to the classic conference the original poster described.

The concerns raised about 'Improv' conferencing are quite valid, but this is not an either-or situation: I have attended and hosted off-the-cuff roundtables that were 'Improv' by almost any measure, then later in the day given a 90 minute presentation with a 17 page handout, two demonstrations and a physical model. The real difference for both was an audience that wanted to be there ("Vote with your feet" is a standard rule of unconference) and my willingness to do the prep work for the second conf 'on spec'.