I was just reading all the threads about mining basically being SHA256, and something came to me. Having written a (mostly) working SHA224/SHA256/SHA384/SHA512 implementation myself, I'm intimately familiar with how it works. Now, to create a valid block, we just need a block header whose hash is under the target, correct? This means we can skip some of the last parts of SHA256, that is, computing the rest of the hash, which doesn't really matter. Now, I realize that this is really tiny, just a few adds IIRC, but even a clock cycle per hash is precious when computing so many of them.
Not sure regarding BTC miners but LTC ones already applied such a trick.
