Aww... I was so excited that I had an idea, I forgot to read everyone else's suggestions. Also, I'm not good at reading C++ and I don't know enough about the existing CGA code to know exactly what the hell you're talking about, so can you tell me if your code is determining the reward before or after the block is hashed? Sorry if that's a stupid question, but that's the only possible problem I can see right now-- if the reward outcome of block 2 is based on the hash of block 1, it's just as easy to predict whether block 2 will pay out as it was before.
Because of how bitcoin (and all altcoins based on it so far) are designed, you MUST know the block reward BEFORE the block is hashed. Every variable reward coin so far is predictable (you know the reward before the block is found).
That's a major problem then... If the block reward must be known before the block is solved, then any attempt to obfuscate said reward is inherently flawed. Again, I don't know nearly enough about how this works to know if this is possible without coming up with a completely new system, but is it in any way possible to rewrite the code to eliminate that restriction? Because, barring some sort of insanely clever idea, that's the only way we're gonna get rid of this exploit.