There was a program I was using that used libtommath, and used that library's mp_rand function. libtommath's mp_rand does various trickery but essentially adds two results of rand() together[0].
If you call rand() multiple times on OpenBSD you will get a sequence that alternates between odd and even, which is completely standards compliant. Because of this the result of adding two rand() results together will always be odd (odd + even = odd).
The program didn't need cryptographic levels of randomness, but only returning odd numbers turned out to be a bit of a problem.
[0] https://github.com/libtom/libtommath/blob/develop/bn_mp_rand...
A Monte Carlo simulation absolutely needs to be able to produce the same numbers every time if you want to have a hope of debugging it. In a release version, yes, sure. But even there, people will still have concerns about replicating other people's results, so being careful with random seeds is important.
If you need high quality randomness, you probably should use urandom. If no where else, when the program is first executed.
And if you're on Windows: https://github.com/php/php-src/blob/e6ea376a91514c59d21f4b53...
Just use urandom. Or getentropy() if your OS supports it.
If you're not using it for cryptographic purposes, then I don't see why it matters. :)