This is an interesting post , though I have not read the paper. I think the blog misses the point a bit with the suggestion that all Weibo posts could be monitored before publication. This would probably break the dynamic of the system, as other news channels would suddenly be faster. Still it is a good writeup of an ongoing problem. There is a lot of ongoing activity (not only from China, but also in the West) of trying to manipulate the information spreading on message networks such as Weibo and Twitter.
The idea of having manual censors using only technology looking for keywords, as described in the blog post is probably rather naive. I think (and perhaps the paper deal with it) that propagation path signatures are applied (did anyone do research on this?). Basically, not only the speed by which the message propagates but also who re-sends it will give a very good idea about the content, and its potential. Remember, if you want to censor something primarily care about its potential, and later look at exactly what it says.
Within a network you may locate nodes (individuals) who will, with a high probability, propagate a certain type of topics, and those who will seldom do so (while still receiving messages "early"). By looking at a group of such nodes it is probably possible to classify message types, and filter out "offending" messages.
Not a suggestion, but it would surprise me if this is not already happening.