Enabling the plugin trace you can see the error
- TRACE: lib_rio: wait_on_socket: ServerIOTimeout fired.
In my case the IBM HTTP is on the same machine as Portal and it's not a cluster. The default value for ServerIOTimeout is -1.
Ajusting to ServerIOTimeout=120 solve the problem
You can see in the TN 1411685 that this problem was found on System I but my server is on Linux.