This issue is part of a broader discussion that is still ongoing in the Dublin Core and SWAP communities. And it will become increasingly significant with Object Reuse and Exchange (ORE). The actual thing that ties all the different bits and pieces together from various other sites and databases may be nothing more than an html or xml encoding page. But that actual thing or encoding page is nothing but a series of coded commands to collect stuff from everywhere else and to display it all in a certain way. That encoding page is the closest thing to the actual resource. I think of it as the "central nervous system" of the bigger thing. The bigger thing, that is, the display that users see and use, is not what is harvested, nor even possibly actually identified. And it does not even really exist — except when that central nervous system is activated. What is stored and manipulated and edited and used in a database will not be, then, what is actually seen by users.
So we are looking at the concepts of technical FRBR manifestations that are not manifest, of resources that are not real, of resource entities that may not really exist — except as an xml datastream linking to a host of other uri's and a series of commands for certain software to display it in certain configurations — and even those display configurations may well be variable, never constant. For example, is an xml datastream might be viewable as pdf or html or other, and the user has the option of deciding this without ever actually seeing the "original document", which is the xml datastream.
The answer is not with us yet. We don't know what it will finally be. It may well be that RDF and ORE — and the Semantic Web — will force us into new ways of conceptualizing "resources" and "resource types".
In the meantime, I would think the safest option would be to store and describe etc webpages the way LOC's Minerva does (www.loc.gov/minerva/). And then be prepared to mark such collections with a sign, "Watch this space!"
Neil Godfrey