After seeing Avalon's gigantic power supply design and huge heatsinks, I start to worry that the BFL products could not be easily run at advertised frequency (Maybe jalapeno and small single can do)

I remember that someone discussed before, going from a 110nm process to 65nm process in principle won't generate a 80% increase in efficiency, the best estimate is around 50%. There was a possibility that Avalon had an unefficient design, then BFL might gain another 30% efficiency by using a better design

But now I have closely examined Avalon product, I think their design is already very modulized and efficient (although not very delicate, but carefully planned to reduce the loss of power), BFL won't gain too much efficiency by a better design

Take these numbers into count, each BFL chip roughly equals to 1 avalon small unit (10 chips), containing 16 of much smaller sized engine, and that single chip will require similar amount of power supply that 1 avalon small unit requires (about 25W), providing 50% more hash than avalon small unit (2.75GH), e.g. around 4 Ghash

small single 4x25 W = 100w
single 8x25 W =200w

Suppose these can be further trimmed down due to 65nm's lower heat generation, let's say that 120W would be needed for a single, then the current power supply design on the board will not be sufficient, if you run it at 60W, then the hash rate will drop by half

So I guess BFL might face a decision of changing the power supply design (which will cause months of delay) or lower the claimed performance and get customer some refund/discount, or any other solutions in the middle