aes performance (was re: public key size)

If AES256 does not slow down the operation of the system significantly
more than just using AES128, then I can see no reason not to use AES256.
 Anecdotal evidence of one implementation on one machine is good, though
I'm finding conflicting results published out there.  Two papers are
attached - one, from NIST, saying Rijndael is weird in that encryption
speed actually *decreases* with larger keys [page 5, aes-java.pdf]. 
The other, an oft cited AES benchmark report for java, has 128bit Rijndael
encrypting 30% faster than 256bit Rijndael (with the same 30% slowdown
for decryption).  Key setup for encryption and decryption had a slowdown
of 36% and 40%, respectively.

All of these metrics were performed on approximate NIST hardware [pentium
200, 128Mb, NT4.0]

The python lib you're using may implement 128 and 256bit in essentially
the same complexity, or there may be some variance in the benchmarking
methodology.  From the details shown in these papers, the answer is not
clear as to whether or not Rijndael 256 bit is significantly slower than
128 bit, as the former says it isn't, while the latter says it is.

How about we go with the easiest for the time being, as we can replace
the session key size later on during peer review.  

thecrypto: which would be quicker to get fully operational and committed
- AES256 or AES128?

-jrandom