I confirm that async=false fixes the shout2send pipelines on all of my bugged systems.
That explains what got broken after GStreamer 1.14:
shout2send now uses non-blocking I/O and has a configurable network operations timeout.
as I guess that async=false restores the old synchronous behaviour. However, that doesn’t yet explain why the issue occurs also when using encoders in the pipeline even without shout2send.
The async=false workaround serves my purpose for now but the more I investigate the more I feel like something stinks on the latest GStreamer.