I meant to bring this up with Charlie Kaufman and Radia Perlman when I got the book, but never got around to it.. The following line of thinking was inspired by the NSA "patch" to SHA: the addition of a rotate-left-one-bit operation to each round of SHA, which yields faster mixing between values in bit positions within the input.
The quick proof of why the initial and final permutations add nothing to the security of DES is presented in an informal and quite obvious sidebar.
Actually, I found the proof not completely convincing. It's really a proof that it doesn't strengthen the DES against a *known plaintext* attack, but there are other weaker attacks it may make marginally more difficult (or easier..); for instance, partially-known plaintext or some forms of verifiable plaintext. The FP/IP don't add *much* security, but I'm not about to say that they add *no* security.
The comment is made that by the same argument the permutation of the key in the generating of per-round keys also adds nothing to security.
Again, their proof assumed that all the bits of the key are independant and unbiased -- which they will be if you're careful, but won't be if you're stupid. Remember that the NSA designs ciphers for use by people with a high school diploma or less :-). - Bill