Strangely, running this newly added use case on its own succeeds every time, and occasionally fails when run with all the use cases.

Ah, I understand. Since the total number of requests is accumulated after the response is completed, it seems that it is impossible to accurately limit the maximum number of requests at present.

def on_response_complete(self) -> None:
    # Callback for pipelined HTTP requests to be started.
    self.server_state.total_requests += 1