Hi,
yes, Bradh is right, one could only make a short resume of both methode, i.e subspace methode is only effective for small eigenvalue problems and for computer with small size of memory (but think about the memory size of computer todays) and lanczos is effective for a great problem and small one too, besides it is robust, since it implemented both "tranformation" and "tracking" methode, so it should be preferebly over other methodes nowadays, unless new methode is invented ;-)
cheers