See this thread for some hints: http://forum.kde.org/viewtopic.php?f=74&t=117721&p=293609#p293609
A proper patch is welcome.
I would also like this feature. However I think it would be worth it to implement the distributed matrix interface, especially in terms of parallel efficiency. However, this requires a loc2glob mapping as an additional function call to pastix, as well as some reordering of the matrix and will likely require some MPI calls inside the wrapper.
If this is something you think will be useful to have in Eigen then I would have some follow up questions about the design.