Fastest, for now, is mine:
http://www.cs.cmu.edu/~dga/crypto/ric/I should be pushing an update in a few hours. Dev fee is 7% now as discussed earlier, will be chopped down and open sourced with no fee within a few days. Only xptMiner is compatible with ypool at this point, to my knowledge.
Aha, thanks for your swift reply.
Your compilation problem is because of an STL issue - you may be missing a -I to where clang has its STL, or are using an out of date version.
If it's any help for debugging, the same is done when I use g++4-9 or icpc instead of g++ (it's 4.8 by default in my system).
Are you on a mac? I haven't tried getting xptminer built on mine, but I could give it a go if that's why you're having troubles with it.
Nope, I'm on a linux / wolfdale setup, just looking to optimize through compilation with gcc-4.9, icc, clang, clang++, g++-4.9 and icpc.
The codebase you're using is about 8x slower than the binary release of mine. No amount of compiler help will make it worthwhile. But figuring out the issues you're working on will help when I open source mine - which is going to happen a little sooner than I anticipated.

Stay tuned.
icc, in particular, might help a fair bit, but the bulk of the time is spent in GMP -- so if you want to do yourself a favor, get a super-tuned install of gmp.
I did throw a little ispc at it, but I've removed it for the open source version because (a) I optimized around the need for what I'd been doing; and (b) I didn't want to complicate the build.
There's a nice place where you could throw some ispc in there, though, if you feel like it. But it needs some architectural changes first.