I think it could use some expanding upon. First, as was already mentioned, the index needs to be broken up into smaller topics. Next I think there should be more discussion (tutorial) on customizing the library.
I agree there is little docs on customizing the library. The reason is I underestimated interest people will take in that part of the library. Also, customization support was added quite late in the development. I probably thought too much along the lines of my own use of the library - I never needed to use the customization features.
* What is your evaluation of the potential usefulness of the library?
Great! My opinion is that this library is going to see far more usage than just what it was intended for. My initial reaction when I looked at it is that it's more of a generic hierarchical tree, moreover, once you've populated your tree you can save it off and come back to it another day. Just taking those two aspects into consideration opens the door to a lot of possibilities, far more than just reading configuration data.
In some of my past projects I used it as a primitive serialization library, where data files are human editable (and creatable). This is hard to achieve using boost::serialization, even with XML archive, because boost::serialization serializes lots of housekeeping information, which is in not easy to grok by human (like class-id integers or tracking ids). Another use of ptree was in a program where user interface was communicating with the "working" part of the program by means of property trees. This allowed running the "working" part alone (without the interface) - it was transparently reading ptrees from the disk instead of getting them from interface. Thank you, Marcin