I don't have time to play around with it much because of work, but would it be worth it to optimize for processors without AVX2, but with AVX? The computers I can throw at this coin aren't new enough to have AVX2 available. Or does the AVX2 build have code paths for different instruction sets, with detection of CPUID or similar?
The AVX2 (Haswell) instruction set is only good for multiple-precision arithmetic and SHA-256 as far as Riecoin mining is concerned. Apart from that, there is no noticeable difference in performance. You will have to implement your own large integer arithmetic using hand-optimized assembly for x86-64 which requires years of experience. You should achieve about a 30 percent increase in mining performance with good optimization techniques.
http://software.intel.com/sites/default/files/m/f/7/c/36945