I suggest you to disable automatic sweeping and schedule the sweeping
operation to some time like 03:00am.

This can be achieved by

gfix -s databasename.gdb

For unix systems, you can use crontab
For windows servers, you can use the "AT" command , or scheduler..

I face some problems like you have. I have two firebird servers. One of them
needs a reboot nearly every 10 days, because the performance problems. After
rebooting, it gets fast again.
The other one works properly, it runs without any problems for many months..
The database is the same, they are both redhat linux machines.

I'm thinking of re-installing everything with the latest linux kernel and
RC2

Did you check you interbase.log? Does it report some errors?