As mentioned by another commenter, the bug is that it's (for python) documented to work for 2^64 elements, but "only" works for 2^49. 2^49 is still pretty big... note that for 64-bit integers, 2^49 integers is 2^(49+3)=2^52 bytes... or 4 petabytes of raw data. Even if you're sorting single bits (one and zero) it's quite a bit of data to chew through.
[ed: Hm, that's not quite right. A 64-bit integer is 2^6, so that should be 2^(49+6)=2^55, or 32 petabytes). I think :-) ]
[ed: Hm, that's not quite right. A 64-bit integer is 2^6, so that should be 2^(49+6)=2^55, or 32 petabytes). I think :-) ]