I'd like to understand how this differs from using more known tools such as: https://gist.github.com/brutuscat/2519840

That uses Tor as a forward proxy (rotating for scraping for example). OP uses it as reverse proxy, I.e. Hidden service.

Question: it seems to me like you could get a long ways with a hidden service simply by placing the same key on multiple servers with disparate physical locations. Wouldn't this provide a sort of randomized load balancing, since different connections would find different servers with the correct key first? I can definitely see the need for more sophisticated load balancing if your usage gets large, but for smaller services it seems like this would give a good amount of scalability.

I'm not well-versed in how these work though, so I'm curious.

As far as I know, you cannot have two servers with same key.

It's possible, but it won't do what devishard expects. http://tor.stackexchange.com/a/24

That's very interesting. Do you have any insights into why the descriptor differences are like that?