One of our clients just called because the iAMP server which runs on their IBMi had taken over the CPU! The problem was due to their ending of the TCP/IP subsystems, this caused the iAMP jobs to go into a loop with the message “[error] (42)Protocol driver not attached: apr_socket_accept: (client socket)” being logged into the error_log constantly. We looked around on the net for any similar issues and found that a similar problem had been logged by IBM when the SSHD server had been configured to run in its own subsystem and the TCP/IP services had been ended and a save started. The IBM solution was to restart the subsystem once the TCP/IP services had been restarted. There are also a number of Apache related problems which still seem to be outstanding but they are more about the graceful ending of the server.
So if you are running ANY TCP/IP based programs that could fall into this category make sure you end them before ending the TCP/IP services. This might be overkill because we also notice that some other TCP/IP programs which are running sockets did not display the same effects but its probably a better option to end all of them than trying to find out exactly which programs are effected.