My initial reactions are mainly two things:
1) The time it would take to create these systems with quality cases that help students focus in on the skills and content knowledge that you want them to learn would be, or could be, immense. The time that it would take to create a whole database of information that was easily accessible and easy to use would probably be four or five times the amount of time it would take students to use the database for a lesson. This is probably not feasible for a single person and would be difficult to implement at the k-12 level.
2) Student self-direction could potentially be a huge issue. These kinds of learning model assume that students already have some kind of analysis skills that will allow them to study the cases and have some meaningful takeaway. CBR and cog flex assume that they will be able to analyze inconsistent patterns over large amounts of material. As someone who is at the graduate level, it was fine for me to go through and complete the task. I also found myself picking out patterns that had nothing to do with the topic. I teach mostly fifteen year olds that have barely been able to choose their own topics to research in class, let alone direct themselves through a bunch of case studies, cross-reference works and pick out relevant patterns to answer open ended questions across 30+ letters. I think this could be good for higher level students with some sense of self-direction, but could be difficult to implement.
I would consider using these models in my classroom if I had a capable Honors, AP or IB class (in which document-based questions are required anyway), but I think it would be difficult. I would have to use a lesson plan that is already created. I might consider using the Plantation Letters in an Honors level class, at least as a Learning Object, but possibly in the cog flex method it was originally intended for.
I would consider using some of the following sources for creating databases of material:
North Carolina Photographic Archive: http://www.lib.unc.edu/ncc/photos.html
UNC Digital Collections (UNC Wilson Library used to be the state archives): http://www.lib.unc.edu/digitalprojects.html
Newspaper archives (there are ads on this one, but the material looked good): http://www.freenewspaperarchives.us/
I would also consider using Wridea in a CBR or Cog Flex model. It allows students to type in their ideas or notes and see them all together. The part I think would be most helpful is you can create "idea rain" which drops the ideas in random orders onto the screen so that you can see all of the ideas together in random patterns. This could be used as a randomizer for the cross-referencing mechanism found in the cog flex model.