We do something like this with our production digital library system,
including over 30,000 images, documents and audio streams. In fact,
the only actual content files that we import into Greenstone are text
transcriptions for full-text indexing. All the images and document
files are referenced in metadata with URLs that point to the content
in a repository system...which was for years simply a Web server with
structured file system under the document root, but now we use DSpace.
I look at the whole thing, Greenstone for Web publishing or access
and DSpace for management and preservation, as the digital library.

-Don