Our finding aid for the Ava Helen and Linus Pauling Papers is currently at 13.8MB of XML.
From very early on I put each series into it's own XML file. They weren't intended to stand on their own so there was nothing "above" <c01>. There wasn't a specific link to them, and I just modified our stylesheet to pull them in where appropriate. Last year we switched to using XML's external entities referencing local files to "link" to the series and are happy with the results. See http://www.javacommerce.com/displaypage.jsp?name=entities.sql&id=18238 for more information on XML's entities.
For web delivery, we have always split the display of the finding aid into smaller pieces. We generate static HTML files and divide the series and box listings into smaller chunks for ease of navigation and retrieval. There is also an option to view the entire series in one file. (The 17 series pages total about 16.4 MB of HTML. The hundreds of smaller pages combined would have a greater total, but most of that is overhead of duplicate navigation). The majority of our traffic comes from search engines, so we've tried our best to make our content easily indexable. http://osulibrary.oregonstate.edu/specialcollections/coll/pauling/index.html
On another note, in 2006 we published a print version of the Pauling Papers. This included some additional content but the entire package ended up being 1800 pages in 6 volumes. http://paulingcatalogue.org/
Mark, thanks for posting about UNC's Hugh Morton collection, I wasn't aware of it before.
Information Technology Consultant
Oregon State University Libraries