I’m guessing it’s because when you are multiplexing by definition you have a shared resource (the multiplexed channel), where you will then need to figure out how to handle access contention between your pool workers, liable to be a pain in the rear, inducing performance bottlenecks, and also, very likely error prone, even in a safe system like BEAM.
If I’m not mistaken, that’s not saying that you can’t “meta-pool”, that is - You could probably do one-pool-worker-per-channel and have the worker do the demuxing and handling of each multiplexed resource within the channel. You just have to think carefully about what the domain intrinsics are and how those partition into pool workers.
