You'll either have to update line 265 of
ITK/Code/Numerics/FEM/itkFEMLinearSystemWrapperVNL.cxx
to use the operator*() method of vnl_sparse_matrix<T>,
or make the mult() method of vnl_sparse_matrix<T> public again in your copy of vxl.

Methods mult(), add() and subtract() of vnl_sparse_matrix<T> will revert to "public" in the next vxl release, be it with a deprecation warning.
-- Peter.