The big question about URL construction, based on the articles linked to in the previous post, seems to be this:
Should URLs convey meaning in and of themselves?
Everyone agrees that they should be short, human readable, and permanent. (Although very few make that hat trick.)
However, the content-neutral approach seems to be at odds with user behavior in decifering URLs to guess page location, primarily by going to the directory level.
Users should win that debate, shouldn’t they?
C. David Gammel, CAE 
This debate isn’t really complete without reference to “INFORMATION-FREE IDENTIFIERS: A KEY TO FLEXIBLE INFORMATION SYSTEMS” parts 1 and 2 by Walter Woolfolk and Bruce Johnson:
http://home.att.net/~Flexibility/Identifiers/Part1.htm
http://home.att.net/~Flexibility/Identifiers/Part2.htm
“The general idea seems to be that as long as the basic requirement for uniqueness is satisfied; an identifier is a convenient vehicle for conveying other information about the thing identified. This destabilizes the identifier and gives rise to the problems discussed in this article.”
The issue here seems to be that permanent (that is, truly unique and truly persistent) is not very compatible with human readable or usable.
URIs are identifiers. Identifiers and labels are not the same thing.