@azonenberg I'm not sure what the best algorithm was back then, tbh.
I'd say at least a factor of 100 in that speed up is parallelization.
I factored a 155 digit RSA modulus in 2012 on my personal workstation in about 3000 core-hours on an i7-2600 (a quad core, so about a month wall clock time), and a 115 digit RSA modulus took 20 hours.
This was with general number field sieve.
RAM is also a significant consideration - this used gigabytes of memory.