d-o-e
... this kind of processing is limited in its application (a) to scenarios where file-writing is part of the pipeline,

Generally: writing to a stream (character or byte stream).
However, there is no real standard for passing d-o-e instructions
(or any other serializtion instructions) downstream, which means
it is best if the XSLT processor serializes itself.

and (b) in the engines it will work on -- i.e., it's not as portable as XSLT generally.

D-o-e is available in any XSLT processor capable of writing to a
stream I ever saw.

The big problem with d-o-e is that it may suddenly stop working
even if the *very* *same* XSLT processor was used. People rarely
recognize that adding postprocessing on a nonserialized form of
the XSLT processor output matters unless they get burned.