OK, PG is dying on that query due to the same bug uncovered in the 3.2.5 datamodel loading problem. Tom Lane, much to his shame and embarassment (his words), broke PL/pgSQL so functions that do a select and find no rows crash PG. He shouldn't feel ashamed, this is the first time I've seen him make a dumb mistake like this in at least a year! At least one that leaked out to release, that is.
The regression tests didn't check this case, so it didn't get caught before release.
