For the final delivery of the transformed or untransformed messages to the destination systems, it's not a bad idea to have one thread for each one. This is because each final system could cause delays. For example, you have a lab system that is configured to only work in synchronous mode and takes five minutes to process each message. It might only get ten messages an hour, so the speed is adequate but you do not want the rest of your delivery service to wait while that receiver does its thing.