May I ask out of curiosity what you would like to use those integers for? Eg. block cyphers, message digests, graphics, physical modeling, amounts of money?
What seems strange to me is that you want more performance than you can get using any of the generic big integer modules, yet you still want to do individual integer operations in perl, as opposed to using higher level functions written in C, such as a cryptographic library or a library for numerical vectors.
For reference to visitors, some (but not all) arbitrary precision integer libraries on CPAN are the two fast backends of Math::BigInt (Math::BigInt::GMP and Math::BigInt::Pari), Math::GMP, Math::GMPz, Math::Pari. I believe that none of the integer vector modules support 128 bit integers: PDL just doesn't, whereas Math::GSL wraps C libraries (GSL and indirectly BLAS) which only support real and complex floating point vectors.
Update 2011-03-18: continuing the list of arbitrary precision arithmetic solutions, see also Convert big number from decimal to hexadecimal.