http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/SFMT/index.html#dSFMT
You might want to use a better algorithm

Do you want to provide a patch for Python's random module?

PS: Please do not upload files that can easily be downloaded from a 3rd party site. It just fills disk space on our server and doesn't provide a benefit for us.

The bar is rather high for changing the current algorithm which is well studied, has a long period, and is reasonably fast.  I do not recommend jumping straight into making a patch before we get agreement that we really want another PRNG when the current one is stable, well-studied, performant, and has been working well for its intended purpose.

Note, we guarantee that the current method will always be available, so this would be just another option.

If we were to add another generator, I prefer that we add one with some completely different characteristics (such as being cryptographically strong).  I don't want to take the default generator and periodically switch it out with the "flavor of the month" (people are always coming up with more PRNGs that make various trade-offs between speed, size of stored state, period, etc).

> If we were to add another generator, I prefer that we add one with
> some completely different characteristics (such as being
> cryptographically strong).  I don't want to take the default generator 
> and periodically switch it out with the "flavor of the month" (people 
> are always coming up with more PRNGs that make various trade-offs
> between speed, size of stored state, period, etc).

Agreed with Raymond.