Ticket is http://hackage.haskell.org/trac/ghc/ticket/3126 .
Sorting by constructor tag is perfectly safe when done right. You can read about how to do it in my 1985 FPCA paper or in Simon's book.
I did, long ago. I learned functional programming by implementing a small functional language, using the Kiel Reduction Language (remember that one? It not only took programming with functions to a level not yet available with modern implementations, it also had a pattern-match sublanguage and engine that was as complex as the rest of it taken together, so we were encouraged to read up on pattern matching in general), C, and your papers. Which made this discussion interesting, as I'm easily intimidated by strongly expressed opinions from people who wrote about this stuff when I was still learning about it;-)
When pattern matching against against things that that are not constructors (like literals etc) it's much trickier to reorder them since you have to prove harder pattern commutation properties.
Good that we now seem to agree on things. Simon: there aren't really any patterns to combine in the test case, so I assume the reordering happens when "combining" a single pattern? Fill the fix you envisioned also cover the IsString variant? Claus