Python has a reasonable efficiency up to 1000 decimal digits (according to benchmark) which is already really great! Operations with more than 1000 decimal digits is an uncommon. IMO you should use a dedicated library like GMP for that.

I would prefer to keep CPython code simple enough to maintain. Objects/longobject.c is already 5744 lines of C code. longformat_BZ.diff adds around 700 lines of code, but it only makes str(int) starting at 2000 decimal digits.

Yeah, we could do better for integers with <many digits>, but I would prefer not to :-) Python has a limited number of volunteers working on it, and it's not specialized in numbers. We should keep the maintenance burden at an acceptable level ;-)