This issue was identified as being caused by load balancers on the servers, because no processes longer than 5 minutes were completing.  It was resolved by changing inactivity timeouts on the servers to be longer for HTTP and all other TCP traffic.