Thanks, it makes much more sense now. However, I can't see why different \epsilon regularizations should give the same results.
There are certainly many different fast-decaying functions, with which we can multiply the integrand, integrate, and then take the limit \epsilon\rightarrow 0 ...