I wrote an entire BPM run-time automation in pl/pgsql. The BPs were stored as PG rows, the deciding data was in the PG DB, and PG triggers automated quite a bit, so the firt seemed perfect. The strong static typing was annoying (I am a Lisper) but did have the usual payoff: if it runs at all it is probably right. Debugging over a socket was rough, but I could take the failed query and drop it straight into a pgAdmin query and get back a very clear error. Guy building the BP front-end picked up what I had done in a few days, and he had never done pg/plsql before.
It took a week to code up, another 2-3 for expressing the initial huge BP and a random BP test driver. Worked great.
My only question is whether I would do it again the same way. I liked that there was not some external system that had to be running for the BP to run, but it is not hard to arrange that and then one gets a much more accessible programming environemnt (such as Lisp). I guess I will not know until I get to do it again.