Everything I've seen so far from Perry on this thread seems sensible
to me.  I'll also note that "randomness" suitable for cryptographic
applications (e.g. IPsec, RFC-1948) is NOT the same as "statistical
randomness".  Anyone who decides to go work on a /dev/random or equivalent
needs to ready/study RFC-1750 first.

I'll also observe that my experience leads me to suspect that any
useful randomness will need to derive in part from hardware (cheap
floppy drives can work well when used in a particular manner).
This means the /dev/random code probably is intrinsically specific
to the particular NetBSD port it is used with.  If the Ted T'so
API could be preserved, this would be a Very Good Thing for 
application portability.

Regards,

Ran
rja@inet.org