That's rather unuseful as an answer.

I don't only want the structure of the site (I have written my own
Sitemapprovider to deal with that).  I want the structure of the pages.
I can manually specify that structure somewhere but that would just be
repeating the structure that is already in the pages.  In order to be always
up to date I want the pages themselves to be the only source for this
structure information.  This means a webservice should be able to
instantiate them and walk the trees (not only the site tree, also the
controls trees inside each page).  I could parse the pages myself but
ASP.NET already has all the functionality I need (and better then I would
write it).  So I just want to access that process in a simple way.

Willy.

Thanks, that put me on the right track.  I was able to use your solution to
get to the class of the page (I also tried  reflection on
System.Web.Compilation.BuildManager.GetCompiledAssembly as well as
System.Web.Compilation.BuildManager.CreateInstanceFromVirtualPath) which all
gave the same results.  Unfortunately, the controls collection of these page
instances were empty (Probably because these instances need to go through a
normal page lifecycle in order to be "filled".)

I finally managed to do what I wanted through the ParseControl method of the
page class (I read the entire ASPX file in memory and pass it to this method
to receive a hierarchical control tree)

I have some work left but mymain problem is probably solved.
So thanks everybody for helping.