Hello together,
I am really glad to have found this discussion!
I think that having a central repository with a dedicated page for each citable item (and possibly subpages for aspects thereof, like figures or tables) is the way to go.
http://openlibrary.org/ are heading in this direction (though only for books), and I have long wished to see some wiki version thereof, preferably with semantic integration. Acawiki is the closest I have seen so far, and if WikiPapers goes beyond that, I would also appreciate the possibility to take a closer look at it. Where can your proposal to the Foundation be found, Brian?
Aiming at systems allowing for two-way citation is a good idea, too, and I would like to add that http://article-level-metrics.plos.org/.
Some further points that come to mind - I have no idea, though, how far they have already been considered in your round (1) the name of the page: Acawiki currently uses the article title, but this creates problems in cases like http://www.ncbi.nlm.nih.gov/pubmed/14707297 and http://www.ncbi.nlm.nih.gov/pubmed/20171346 . Better would be a system based on universal identifiers like DOI and ISBN, if such exists for the kind of media referenced.
(2) Aiming at systems allowing for two-way citation is probably a good idea, as they would allow some meaningful addition to the emerging range article-level metrics that more and more publishers (especially those in the Open Access world) are setting up (cf. http://everyone.plos.org/2009/12/09/article-level-metrics-presentation-to-be... ).
(3) In principle, wikification does not have to be restricted to sources, and semantic techniques could allow to include authors as well (where identification problems are even worse, though solution attempts are on the way, e.g. http://www.orcid.org/ ), or even institutions.
(4) Like with other reference management systems, integration with citation workflows is crucial, e.g. portability to and from BibTeX files (cf. http://feedback.mendeley.com/forums/4941-mendeley-feedback/suggestions/28612... ), or direct links to PDF, HTML or XML files the individual wiki user has access to. Several reference managers are building a very large database of metadata - in part taken from public repositories, in part from PDF indexing and in part from manual editing. There might be room for synergies.
Cheers,
Daniel