There are two things to consider: how much security you want and how much should it cost (money-wise and speed-wise).

But, the disclaimer first: no matter how much security is in the algorithms used, it's completely useless if you use it in an improper way. The absolute majority of security breaches has nothing to do with "DES-cracking" and similar. And, unfortunately, there is NO universal way of telling, whether your application does or does not have security holes; this is a matter of knowledge and experience in the field (the same as programming).

When it comes to commercially available '51s, you also need to consider the "commercial crackers" issue. Most of the commonly available microcontrollers' FLASH contents "reading" is offered by them for an amount in the order of $1k-$10k. Of course, it depends on the application, whether the mcu is made available for such "investigation", measures can be taken against that.

Btw., the free stuff at efton.sk, that's me. The PC counterpart is usually available freely on the net. '51 TEA and SkipJack are in the code library here too; and I see a DES there but that's not mine and I did not test it. I also have implemented blowfish (to be published there and here soon, I need time to write about it a bit), and commercially DES and AES128 (the latter performs slightly better than what the authors published in their paper - around 3000 cycles, it converts to around 15kByte/s in the '51 configuration mentioned below), and a handful of proprietary encryption algorithms which are useless for general purpose. I don't say I am a cryptoexpert, I am just using and implementing these algorithms. And they are all symmetrical ciphers, if it matters at all.

You have figures on the mentioned site for net encryption speed, for most of the algos it's in the order of several kByte/s on a 6-clocker at 20MHz; so you'd be able to feed a 19200bps or slightly faster line with it, more or less, depending on the overhead of data manipulation.

Jan Waclawek