I'm considering encrypting some small files (hundreds of kb apiece) using the AES Crypt reference implementation. Looking at the source, it seems that the encryption key is derived from the IV and the password by concatenating the two into a buffer which is then repeatedly hashed. So the IV is acting as the "salt" for the hash. By repeatedly, I mean 8192 times.
I understand that the benefit of this is to increase the time required to generate the key, making it more expensive to perform brute-force attacks to discover the password. I also understand that the drawback is that it takes longer to do the legitimate encryption and decryption tasks for the real users. Further, as both the users and attackers buy faster machines over time, the benefit and drawback will tend to zero.
So my question is, given current computer capability, and assuming a motivated attacker who doesn't own a dedicated cluster, is 8192 iterations insufficient, overkill, or "just right"? Also, have a missed anything in my analysis of this key generation: is there some other reason for choosing such a number of iterations that makes it a good choice?