First I would like to congratulate the team of .NET Numerics, as the resulting work is exceptional and the library can become the most powerful one for .NET framework.
As a finite element method for structural analysis researcher, I would like to make the following suggestions:
1. The use of symmetric sparse matrices is very common in finite element analysis. A symmetric matrix can have almost double capacity of a non symmetric.
2. Factorization algorithms as well as iterative solvers could take advantage of the sparsity and symmetricity of the coefficient matrix resulting in faster calculations.
3. An eigenvalue solver that would find only a desired count of the first eigenvalues of a sparse, symmetric matrix would be ideal for finite element analysis.
4. Large sparse matrices (that would not fit in memory) could be handled in an out-of-core manner.
All above suggestions would be highly appreciated by all FEA developers/researchers.