So wtf... have you forensically analysed the code yet?

By the way, it looks like the main thing of interest is the way he breaks down the calculation of W into multiple stages. (This isn't really surprising; the W calculation was what was driving me nuts and I think the same is true of FPGAMiner.) The design only actually has 2 pipeline stages per SHA-256 round.
Edit: Also, the map and place and route options - ouch!