I think the key word here is 'arbitrary length arithmetic'. I would think this would be best done in assembler on a 8051 and have 'c' call the required routines. I too are curious why you would need 128bits - my poor old calculator won't even handle a number that big!
If you were doing encryption, then that would explain things a little better - expect to wait a little for the 8051 to crunch that one...........