One problem with the implemented solution is that it is impossible to add content or contexts to an existing neuron that you cannot modify. One could imagine doing a catalog search to add content to those specified in meta-data.
A complex piece of content may generate large amounts of content descriptions. An example is an article like this, that would need one component per important section. One imaginable solution to this kind of problem is server-side generation of content descriptions on the fly from their identifiers, using for example XML markup in the article or other means. Quite generally, it is imaginable that the server stores the meta-data in a database of some sort, together with the content it describes. In this case, exporting the content descriptions over CORBA could be a reasonable idea.