DRK started out as CPU only. I'm sure in time someone will figure out how to mine with GPUS.
The algo is nothing like the others.
X11 was nothing like the others.
Actually X11 was
exactly like the others. It's a whole bunch of conventional hashing algorithms (all fairly easily GPU mineable) combined together. Anyone with half a brain would have known that X11 would be GPU mineable, and anyone with half a brain who claimed otherwise was scamming you.
This algorithm is entirely new. I'm not going to bet my life on it staying CPU only forever (or even for a long time), but X11 is a really, really bad example of why it might not.
+1 to all of the above.
At the risk of shedding a professional opinion - I think CryptoNight will be eventually shown to be GPU mineable, but its developers made some really elegant design decisions in it that will have the effect of having a much better CPU-to-GPU ratio than, e.g., scrypt, which it is similar to in many ways. Whether they're enough to make a GPU *less* profitable than a CPU is uncertain, but they were clever - I admire this one much more than the usual clone-a-coin we see with throwing variants of SHA competition finalists together.
In my taxonomy of proof-of-work functions, this one gets its own category: single-sha-type, multi-sha-type, scrypt, momentum, (maybe MMC), riecoin/primecoin, and CryptoNight all offer something fundamentally different between the categories. Within them there's not as much that's interesting -- simply replacing SHA256 with Keccak -- yawn.
