In the case of a near-partition, where one single node is the conduit that links both halves of the network, the network still operates normally. A message from one half will spread normally until it reaches a peer of the conduit node. The conduit node will then accept it, and pass it to the other half, where it will again, spread normally, until it has saturated the entire network.
This is exactly my point. If there literally was only one computer connecting China and non-China, then all messages which need to cross from one side of the network to the other would go through that one computer.
Unless it was set up correctly to be used purely for bitcoin, it would not only transfer the 300 bits per second of bitcoin, but all the terabytes/second of everybody in China requesting any website not in China. It would go down in an instant.
Even if it were set up to be used purely for bitcoin, it would receive connection requests of everybody in China requested any website not in China, and just refusing those requests would cause it to go down.
But I have to admit I don't know the details of routing protocols. How does this computer announce (falsely) that it doesn't have any routes to non-China? IP? BGP? I don't really know the technical details.
Also, if the Chinese government has a way to figure out where this computer is connecting to the Chinese network, they can cut it off, making it purely part of the non-China internet (even if the computer might physically be located inside China).