samba - Jan 2009 - Log partition fills, process hangs... Need help debugging

Greetings List,

I've had a persistent problem that's been going on for a couple years,
through several Samba and Linux upgrades, and I need some ideas on how to
debug this and provide enough useful information to solve the issue.

Here are the symptoms:

1) An instance of smbd spins one CPU up to 100% utilization, and can only
be kill -9'ed.
2) Either because of #1, or causing #1 (hard to tell which) the partition
that the samba log files are on fills completely.  The sum of all file
sizes of all log files does not equal the amount of space on the partition
- it *should* be less than 25% full.  Simply deleting the log files out
from under the running smbd processes does not reduce the amount of used
blocks on the filesystem.  Only killing the smbd process discussed above
returns the used space.
3) Most other processes continue to function normally, except for a few
profile loading errors, or "delayed write failed" errors, since the
log
filesystem (which also has the locks directory and tdbs on it) is full. 
These errors disappear after killing the aforementioned process.
4) The log files say nothing out of the ordinary when this occurs
5) I can make this occur every day or two if I turn the logging level up,
and/or set the max log file size to anything over 5k.
6) With the logging turned down to the lowest level, and the log files set
to turn over at 5k, the system will go for months without issue.

Currently, this system is Fedora 9, Samba 3.0.29 (compiled from source). 
Although, I've had this error occur across several distributions and samba
versions over the past couple years.

I'd love to post log file entries, but they never contain anything
unusual, even if I turn them way up, and cause the error to happen
quickly.

I'm stumped.  Any ideas on what to try?

Thanks,

-John