"Worker exceeded CPU time limit." when pipeTo data without processing

I’m reverse proxying audio data from an icecast server. It’s basically the example code from the docs: workers/learning/using-streams/. After a while - sometimes minutes sometimes hours - the connection stops and the exception “Worker exceeded CPU time limit.” is caught. Why does this happen? The CPU shouldn’t be used here, am I right?

I’m on an unbound plan. The worker is used to select a valid icecast server based on several criteria.