Roberto picked up on a point I meant to comment on and forgot - languages don't need to be threadsafe in order to be embedded inside Postgres because Postgres itself is process, not thread, based. Among other things this widens the candidate languages that can be included, and means that those writing C functions don't have to add thread safety to their list of concerns.
The primary cost of this is that connection startup is slow. In our case, running under AOLserver, we don't care 'cause we're pooling persistent connections and only pay the price once.
Daryl's query would run nearly unchanged in PG, as he suspects. You could minimize the change necessary by using SQL92-standard CASE rather than DECODE, because PG and other SQL92-compliant RDBMS's support it (as does Oracle as of 8.1.6 or so).
