Thanks. I'm still not certain which of 2 scripts caused the problem, but I did forget to call $sth->finish() in one of them. I enabled $dbh->trace(1) (and then 2 :) and couldn't find anything wrong. Hopefully it works now.

This brings another question to mind: under the right (or wrong...) conditions, I might have a SQL query that hangs for a long time. Supposedly, the webserver has a cron job that kills CGI processes after 15 minutes of execution. Is it possible that whatever signal this cron job sends could cause Perl to exit without calling the destroy method on the database handle?