Ross Atkinson argues:

The library also has a responsibility at the citation level - ideally ensuring that all information cited in local publicatons remains accessible. This should be the case especially for electronic information: if Web pages, for example, are cited in publications produced at a library's institution, then that library should be responsible for ensuring that those Web pages remain accessible idefinitely - preferably in the form they took at the time of the publication. [Uses and abuses of cooperation in a digital age. In: Community, collaboration and collections: the writings of Ross Atkinson. Chicago: ALCTS, ALA, 2005.]

How would one do this, I wonder?

If we had a generally accepted persistent identifier approach it might be easier. Maybe it will become a part of deposit in institutional repositories? Maybe it already has: I am not familiar enough with local practice.

Of course, there is a similar issue with data sets in scientific research where databases change over time. This is one discussion of the issue:

Keeping the past states of a database is an important part of the scholarly record. [What the web has done for scientific data - and what it hasn't. Peter Buneman. [pdf]]

The scholarly record ain't what it used to be ;-)

Which means that libraries and research libraries have some serious thinking to do. What is the scholarly record in these changing times?