Please Help - data duplication when server under load

[Bradley Shank]

Hi,
We are running mysql-standard-4.0.21-pc-linux-i686-icc-glibc23 on
redhat 8. During heavy use (and when doing backup using mysqldump)
we have noticed duplicate records (with unique id's from auto
increment) as many as 74 duplicates for one record, in one table about 1%
of the inserts result in duplicates! . The table where
most of the dups appear is type MyISAM and contains over 4 million
records and receives mostly inserts. We are using persistent
connections with a timeout of 20 sec. The duplicate records occur
when the cpu utilization spikes to 99% ( another strange anomaly as we
haven't seen this high of cpu usage from our other mysql servers) The
server is a dedicated db server with 2 GB ram and a 3 GB Xeon
processor. The application inserting into the database is a php
application (php 4.3.8). I have thought about changing the table type
to innodb but we want maximum performance and I think this will slow
things down too much also this is our production machine and we can't
afford any down time. We have binary logging turned on to help
diagnose the problem and the only strange thing is the dup records
have a different thread id. Any ideas???

Thanks in advance,
Brad