SRANDMEMBER's prng kinda sucks

[mattly]

While writing tests for some code that makes heavy use of SRANDMEMBER,
I've noticed that it seems to heavily favor certain items in a given
set.

Here's some examples using the ruby client: http://gist.github.com/607299

Is this a known issue with SRANDMEMBER ? Our application requires a
more even distribution.

Thanks,
Matthew Lyon