I haven't had a chance to run it through the DieHard tests. I've only looked at some preliminary statistics. And Streety, the throughput is about 19200bps. With the Von Neumann filtering the it's halved.
As for attacks with Freon, if the input becomes highly regular, i.e. lots of continuous 0s and 1s, the VN filtering will compensate, though output will slow.
I spent a lot of time thinking of a good algorithm to calculate the median with the limited RAM of the Arduino. The granularity of the calibration is limited by the array size possible in Arduino. If you can think of another way I'd love to hear it.
With more computing power the system could do some self diagnostics and warn of non-random behavior. A previous project of mine did essentially that: