Backend process that won't die

I have a couple of backend processes that are "stuck", and do not respond to a pg_cancel_backend. This is PostgreSQL 8.3.5. The pg_cancel_backend returns true, but the process keeps running. I have also done a "kill 12345" from the command-line, with no effect.

The processes are running a "select function_x" statement that normally takes a fraction of a second to run.

No locks are shown when I do:
select relname,pg_locks.* from pg_class,pg_locks where relfilenode=relation and not granted;

We had a database crash last week, and had to reindex a bunch of tables, but this function has been working for several days on the same tables that should be being used by the function_x function.

Any ideas on how to get the processes to go away?

They are eating cpu cycles, for no good reason:
postgres 28396 85.0 1.4 4420768 242224 ? Ss Sep03 3193:40 postgres: userxx dbname1 172.27.43.9(1160) SELECT

Thanks,
Susan