It's a good question, discussed a bit in the PETS paper: https://crypto.stanford.edu/flashproxy/flashproxy.pdf (section 6). The short answer is that communicating a small amount of information outside of the censored region to a blocked facilitator is an easier problem than full connectivity.
Yeah, it was a stupid question in that I should have read the paper ;-) Laziness on my part, sorry.
Right, their idea is to basically leverage some protocols that allow unblocked entities forward information from the client to the facilitator (which is assumed to be directly blocked always). That seems to make sense.
It seems to be that the flash proxy must be able to connect to the client directly (the client is not behind NAT), which seems like a pretty big assumption... though that might be fine in the real world use cases that they are targeting.
Plus IPv6 could drastically reduce the need for NAT (particularly in mobile where they're basically out of IPs and basically NAT the entire mobile network)
