We pondered this observation before, see axn's link and a few other threads here. The tentative title is "extended crank P minus one smoothness hypothesis", "ECPMOSH", or something like that... I forgot. At any rate, if the smoothness of p1 has an effect on the probability of Mp being prime, then I'm fairly certain it is due to the smoothness of p1 affecting the probability that Mp has small prime factors, and thus after trial division, the survivors should be practically equally good candidates for primes again.
