lightweight counterpart for collation()

Hi list,

As I recently ran into storage problems with collation()
I would like to see a lightweight counterpart with the
following features:
- syntax like in collation() for one directory
- on invocation gets a sequence (dir-seq) representing
  the documents in the directory, the order determined
  by the full URI, including the member-name. If the
  function is called again, it reuses the dir-seq.
- each document is fetched only once from directory
  when it is referenced for the first time. It will be discarded
  automatically if an other document from the same dir-seq
  is referenced and it will not be re-fetched later on.
With the Saxon implemented collation() I noticed that the
sequence I got was not ordered by the URI and I had
to use sort, which in terms of memory usage can become
very costly indeed.

Any thoughts will be appreciated

Manfred