In the wake of 77a5e9a6, the performance of the FLENS-dependent parts could be boosted by looking for ATLAS or other optimized LAPACK implementations on the system in cmake, and when found, instructing FLENS by the appropriate compiler macros to use them instead of its own C++ implementations.
Note: my tests have shown that there is no perceivable difference in performance between the generic netlib implementation and the FLENS C++ implementation of LAPACK.