Of course you don't use /dev/random for number crunching. If you
want a really strong random number generator, you'd use /dev/urandom.
OTOH, for Monte Carlo simulation etc., you don't want really random
numbers. Your best bet would be a good (and I emphasize GOOD) random
number genetator (you can base it on MD5, for example), which you feed
with different seed values. Keeping the seed values around is
a good idea in case you want reproducible results (like running
the same program with different optimization settings to see if anything
breaks).
>In fact it was a Montecarlo integration, which made me sceptical
>about /dev/random or /dev/urandom.
The algorithms in /dev/urandom are believed to be very strong. If you
can point out any weaknesses they have in Monte Carlo integration,
please post your results to sci.crypt.research. You'll cause a minor
revolution there.