Well, there is no reason that the value of the DOCUMENT_ROOT can't
depend on the URL, so you could accomodate multiple roots. But
I don't have strong feelings on this. I would be happy with having
it be server-specific.
You should be aware however, that (as someone described recently in
a www-talk post) script writers are taking PATH_TRANSLATED and
trying to delete PATH_INFO from the end to obtain a document root.
They are doing this because their scripts need to access more than
one asscociated file. How will this work with cern_httpd?