John,
In the PEM environment, this issue probably boils down to
whether the 2nd 8-byte quanta of plaintext (or any later quanta) is
significantly less predictable than the first. In CBC or CFB-64
modes, a know plaintext attack can be mounted on any 8-byte quanta is
you know the underlying plaintext, since the pseudo-random quantity
XORed with the Nth plaintext block is just the N-1st quanta of
ciphertext. So, unless there is a big difference between the
predictabilty of the first block vs. later blocks, there does not seem
to be an obvious, significant cryptanalytic difference (said the guy
who knows he's not a cryptanalyst!).
I wasn't familiar with the "suggestions" you cite for
exportable crypto, but in my experience with a pretty wide variety of
crypto, sending the IV in the clear is a common practice.
Steve