This issue, and its history, was discussed at length in Silverman and Rivest. The relevant passage here is in Section 6, which I quote:
In 1977 Simmons and Norris [53] discussed the following "cycling" or
"superencryption" attack on the RSA cryptosystem: given a ciphertext
C, consider decrypting it by repeatedly encrypting it with the same
public key used to produce it in the first place, until the message
appears. Thus, one looks for a fixed point of the transformation of
the plaintext under modular exponentiation. Since the encryption
operation effects a permutation of $\mathbb{Z}_n = \{0,1,\ldots,n-1\}$, the message can eventually be obtained in this
manner. Rivest [46] responds to their concern by (a) showing that the
odds of success are minuscule if the n is the product of two
$p^{--}$-strong primes, and (b) arguing that this attack is really a
factoring algorithm in disguise, and should be compared with other
factoring attacks.
(p - 1)
has a large prime factor requires very little extra effort. – Jul 06 '12 at 08:44