Danny Bickson, Large Scale Machine Learning and Other Animals, Spotlight: Blaze C++ math library, here. The Iglberger, Hager, et. al. papers referenced in this blog look interesting.
Whereas in direct comparison Blaze cannot compete in the total number of features, Blaze still offers a small number of unique features. The probably most important is the support of the Intel MIC architecture (Xeon Phi). Second is the support of the AVX instruction set, that is still not available in most other C++ math libraries. Third, Blaze is probably the only library that allows a completely hierarchic nesting of matrix and vector data types without performance penalties.
So the difference between the Blaze performance curve and the Eigen3 curve, say in daxpy performance here, is more or less AVX versus SSE? Eigen 3 keeps up until the vector length is around 10 and the differential performance is about a factor of two until the cache limitations kick in. I would expect the MKL using AVX as well – so that performance curve seems to merit some followup questions