Vectorz has some interesting properties which I need and are far more important to me than raw performance on stuff like matrix multiplications / decompositions which is what the benchmark checks

stuff like:

- Being to be able to join vectors into concatenated "views" without copying

- Full n-dimensional array support

- Specialised matrix types (diagonal matrices etc.)

- Sparse matrices / vectors

- Views for subvectors and submatrices

- Specialised small primitive vectors, e.g. Vector3 that fit the same type heirarchy

None of the other matrix libraries offer all that, to my knowledge

There's also stuff like first class mathematical operator objects, immutable arrays etc. which may be useful for some people