I realize you mentioned it briefly but I would recommend trying to convert x^y to e^{y ln x}. Of course, this would mean that you would have to compute e and ln to the required level of precision (so that your result would be accurate) but at least your solution would be exact and you would know that you were definitely approaching the solution.

I don't know why you mentioned having to brush up on C/C++ programming? Presumably, the Math:: library is coded in Perl - no? You should be able to add e and ln functions to the package.