hmm.. looks like this may be the problem.  netstat reports 16 tcp sessions
established to 783/tcp, which would indicate all the children are busy right?  
however, there is no load, and nothing in the log ever showed max children reached.

[root@spamd1 log]# netstat -na | grep :783 | grep ESTABLISHED | wc -l
     16
[root@spamd1 log]# netstat -na | grep :783 | grep CLOSE_WAIT | wc -l
    123

so the parent wont accept any new connection because it thinks all the children
are busy, when they really arent?  ps shows they are all sleeping...

15515 ?        S      0:03 /usr/bin/perl -T -w /usr/bin/spamd
--syslog-socket=none -A 127.0.0.0/8,24.225.0.0/24 -i 0.0.0.0 -q -x -m 15
--max-conn-per-child 75
 9076 ?        S      0:10 spamd child
 9127 ?        S      0:09 spamd child
 9152 ?        S      0:15 spamd child
28765 ?        S      0:20 spamd child
29097 ?        S      0:15 spamd child
29139 ?        S      0:12 spamd child
20022 ?        S      0:26 spamd child
20177 ?        S      0:23 spamd child
20191 ?        S      0:23 spamd child
20444 ?        S      0:17 spamd child
20611 ?        S      0:10 spamd child
20682 ?        S      0:08 spamd child
20696 ?        S      0:18 spamd child
20698 ?        S      0:07 spamd child
21294 ?        S      0:04 spamd child

strace shows they are in a read().

[root@spamd1 log]# strace -p 21294
read(6,

is something just not cleaning up properly?  here again is where a timeout on
spamd children would be a good thing.  the parent could sigterm the child if it
hasnt finished in x seconds.

hmm.. not sure how this got on bug 3829 vs bug 3828 where it should have been..
my mistake.

Looks like this is fine now.