There was one attempt a while ago to get mysql to work with arc.
As for using arc as an cgi script, it has a rather slow load time. If you are loading for every request it will be quite slow. That's one reason that arc includes it's own web server.
One common way of using arc is to run it behind an apache server with mod_proxy.
[ I guess opening a socket to the db server and parsing the resultset wouldn't be much harder if its text mode - could probably use the perl or ruby pglib as a guide, and port from that... ]
Hmm. I would think that you wouldn't want to open a new psql connection every time you want to look something up.
How about running the system command in a thread, which redirects psql's stdin and stdout to named pipes so that another arc thread could read from it using file read commands? That would give you more persistence, so you wouldn't have to keep loading psql.
Also, have a look at lib/mysql-ffi.arc in anarki; supposedly it has a ffi for mysql, though I haven't really looked at it much myself.