It's not a rigorous academic study of the merits of Brainfuck as a steganographic technique.
Hackery, pretty much.
> if the message was discovered, then brainfuck does not represent much of an obstacle to reading it.
The point of stego is that the message is not suspected. Usually this is done via statistical analysis(1). However, this just tells the adversary that "a message could be there". Given that information, they still need to find out which bytes are information-carrying, and that those bytes encode to a BF program. Stego is all about the extra hurdles, it is literal security-through-obscurity. But once they got that, yeah, the message will be exposed.
(1) Which is why encrypting the message before stego is not a straightforward good idea, as encrypted data looks like white noise, and contrary to popular belief, the lowest bits in most photos are not distributed like white noise (as a naive stego algorithm would just hide an encrypted message in the image's least significant bits).
This is true as far as it goes. It should be noted, however, that in many threat models for steganography, this problem is solved through rubber-hose cryptanalysis. http://en.wikipedia.org/wiki/Rubber-hose_cryptanalysis
Slightly less awful than torture, you also lose plausible deniability in a legal setting, or even other situations.
I agree the most important thing about stego is that the existence of the message is not discovered at all.
Sounds like that means encryption before stego is pretty much useless, then? Still something in my head says "better safe than sorry", even though you're not mitigating all that much risk.
A better idea might be to compress the message before stego. That way you need to hide less bits, even though the bits will be distributed like white noise, which can be harder to hide, but could perhaps also be something that can be relied on in some sense.