SFS Servers and non-loopback NFS servers

Hi,
   As I understand it (from the documentation and list archives) a
machine can run an SFS server and have it talk to an NFS server on a
different machine.  This used to be acheived using the NFSHost
directive.  More recent versions simply look at which file systems are
to be exported and if any of them are NFS mounted from another machine,
the server talks to the remote machine directly.  This is a more elegant
solution but requires the remote NFS share to be mounted on the SFS
server.

I'm attempting to set up an SFS server on a linux vserver virtual server
( http://linux-vserver.org/ ).  This poses a few problems.  First of all
there is no loopback address, the kernel NFS server can't be used and
the kernel NFS client is to be avoided if possible.  unfs3
( http://unfs3.sf.net/ ) and the old NFS user space server both work and
can export the filesystem to the main IP of the machine.  Thus if the
NFSHost directive could be used everything would work fine, without
needing any kernel space NFS work.

So, 1. how much work would it be / is it possible to re-enable the
NFSHost directive? 2. are there any subtle reasons why this would not be
a good idea? and 3. are the developers willing to do this / willing to
accept a patch to do this / able to suggest a better solution?

I appreciate this is something of a minority case but it would be nice
to be able to run an SFS server on a POSIX environment without kernel
NFS support.  Add a pure userspace client (KIO already supports NFS,
this might make an interesting student project) would potentially free
SFS from any need of platform NFS support.

Cheers,
 - Martin