There's a nice example of explicit multithreading here, but coming from ROS1 .launch files to launch.py scripts we lose the implicit multithreading and thread pooling of individual nodelets in a node manager but instead get everything in a single thread with the associated performance hit.

Am I missing a simple way to add all of my ComposableNode()s to a thread pool or something similar?

UPDATE: I've reported this as a bug.