Who is the library for?
How do you anticipate it being used? (Or how do you think it should be used?)

We've all seen and recommended books for various reading lists. Most of those lists are far too long to even scratch the surface of - and many of the lists overlap with lots of the same books, many of which, imo, are recommended simply because they are popular, often times with the recommender having not even read it himself. They are nice to peruse, but not of much use, imo.

Assuming that a library, as opposed to a magazine stand or a bookstore, is a repository of information organized to help people find what they are looking for, I think it would be ideal if the SWJ library helped to guide readers through subject areas. For example, let's say that I know squat about Iraq. It would be nice to be guided in a way like this...
Here are some good references that provide an overview of the history...
the culture...
the current issues...

Quick primers should be the first recommendations. If the individual wishes to delve into a specific area, then drill down (click on a link) into the longer, more in-depth materials.

It should also be more than just a collection of references. Short explanations of what the individual can get out of the book/article/video/etc would be helpful. I'm thinking about the short blurbs that you see in reading lists at Amazon.com (but preferably a little more depth than three sentences, to include comments by others who have read the books). Allowing users to post book reviews would also be ideal.

Now that the web has matured, recommendations can include not just book recommendations, but video clips (Charlie Rose interviews, for example), blogs by knowledgeable folks that focus on a specific AO (Registan), and well-written material from locals (for example, Iraq the Model).

Lastly, one comment regarding the logical program design (as I now don my amateur programmer's hat). There needs to be a significant amount of thought put into the design underneath the hood - not just the user interface. I am not sure if this means an XML-based tagging feature, a relational database with lots of joins, or some crazy google-esque searching algorithms. That, imo, is where the real heavy lifting will be (and where you'll probably need to spend the most money, unless you can find some highly qualified individual who will do the work on the cheap, if you can win his heart and mind).