Story

I believe that the primary problem is simply waiting for high quality bits of entropy. High quality DRBGs are all about "quality in" = "quality out". Testing the quality of that entropy is actually a bit harder than it seems and in my line of work, we tend to overestimate how strong those sources of entropy are. If you need to generate 256 bit AES keys, then you need at least 256 bits of entropy: which is not the same as typing 256 random keys on the keyboard to get it. You might only get 1 bit of entropy for every 20 bits streaming into the entropy pool -- and this is the real danger of /dev/{u}random, rather than any blocking or non-blocking behavior.