Jez:
Well considering that a 4096 bit number is on the order of 512 bytes......you should consider an "arbitrary precision" number package. It's already going to be slow considering the use of XRAM so go for a C implementation!! Look into the work done by Michael C. Ring. He has published a very nice article on the subject in the C/C++ Users Journal a couple of years ago. You can read all about it at The MAPM Home Page

Michael Karas