Semantics, meta-data... Who is going to enter it?
This reminds me of the new windows search for Vista : You can find any data via a keyword search - but only if you have tagged your files with the appropriate keywords (ok, I'm paraphrasing).
If the search is fully automatic, then it is indexing plain data to metadata, and this does not seem fundamentally different from any other search index.
If the system relies on metadata, then someone either has to enter the metadata by hand, or by... an indexing system that translates data to metadata.
The advantage of this is that the indexing work is carried out on the site side and not on the search engine side.
The inconvenient is that the metadata "categories" are generated by the criteria that are site dependant, and not search engine indpendant, and this means that the search engine would probably have to reindex the metadata to fit in it's own categories.
The other point is that this would seem a good candidate for search index poisoning: The medadata from whatever feed linking to a page could be extracts of a medical or technical journal, but the page referenced by this feed would be of questionable security (porn, botnet installer) etc... What you index may not be what you really get, unless your are reading their site "digest" and not the associated complete page...
Cheers,
Daniel