SuperLU is via Pysparse, CHOLMOD is via CVXOPT. Both are direct sparse solvers, from what I've read it seems as though CHOLMOD might be faster, since it is specifically for positive definite matrices that result from FEA problems. If it's only marginally faster (<10%), it won't be worth it since number of dependant 3rd party packages will remain the same.
SuperLU is via Pysparse, CHOLMOD is via CVXOPT. Both are direct sparse solvers, from what I've read it seems as though CHOLMOD might be faster, since it is specifically for positive definite matrices that result from FEA problems. If it's only marginally faster (<10%), it won't be worth it since number of dependant 3rd party packages will remain the same.