I'm observing some unusual fluctuations in benchmark times for both $powmod and the original scripted bf_modpow alias
That looks fine to me. The only changes to $powmod() were the additions to it from your previous post.
There were no other changes to the library other than fixes to specific functions in MAPM that were needed to prevent freezes - these changes simply reverted previous changes I had made to the library which I thought would speed it up but clearly caused other issues.
Another possible cause for a slowdown is that I increased MAPMs maximum precision setting so that it was at least 10 digits greater than the maximum 30 decimal places that mIRC returns for bigfloat calculations.
At this point, we don't really have a baseline for how fast bigfloat should/will be, since we are still experimenting, tweaking, and fixing issues. Once It looks like it is stable and is returning reasonably correct results, we can think about how to improve performance, although at the end of the day, this will really be limited by MAPM.
