I got around to hacking up a version of Djangos
cache-middleware, which understands and obeys the
response-header. This means it will cache different
copies of the pages, based on what the request headers are.
So, Firefox will get a XHTML page from the cache, while IE will get an HTML page. And, I will get a page with admin-links from the cache, while you will get a page without admin-links.
This is good.
http://mel.ibofobi.dk/~sune/r/ibofobi.git, now I just
need to figure out, if it is interesting enough to
include in the official distribution.