Quoth Joerg Arndt:

Get an FPGA board, implement a (deterministic) random number generator (with _very_ long period), and XOR in

unbiased (Neumann will do, and you can recurse on the asymmetric diagonal if you're desperate for bits.)

entropy from several independent noise sources (on that FPGA board, say 5 such at $3 each ("enterprise solution")). ... Problem solved?  No, people will not use it because they do not realize how bloody important the "random" part is.

Amen! "Random" is a minefield. One of my pet rants is against the ignorant users of the Mersenne Twister, as, almost without fail, they'll seed the MT with sometimes as little as 15-bit of entropy, and then happily assert they're using the best random number source known to man. (Not all users are ignorant. However, after inspecting about a dozen open source projects, I'd claim that over 92% of them are. Worst of all is when they're willfully ignorant, in the face of bugs being filed against their PRNG.) Phil