I believe the current lowest-memory algorithm for computing the $n^{th}$ binary digit of $\pi$ requires $O(log(n))$ bytes and $O(n^2 log(n))$ days (I pick Bellard over Bailey–Borwein–Plouffe for speed).

Can any of the common irrational constants be currently computed with the same low memory, but in faster time? Please consider constants which combine integers and $\pi$, natural exponential functions, and/or root functions (e.g., $\pi$, $\pi^2$, $e$, $e^{-\pi}$, $\sqrt2$, $\sqrt{2\pi}$).

For any such constant, I am also curious about the randomness in its binary tail. These digit extraction constraints cause correlations in the constant's tail bits which probably go to zero for high $n$ (but, for $\pi$, the Bellard/BBP constraints are too subtle for me to conclude anything). If anyone has a low-memory method to distinguish a tail of some common irrational number from random bits (with the same speed in the limit of high $n$), please comment.