That's a very good point. I suppose that, if we assume a bitrate, and we assume partial messages are irrecoverable, then this puts a limit on the "uncensorable bandwidth." But it's easy to imagine a protocol that splits large messages into very small pieces to so that each message is below that limit.