I think 8 bit primes is probably a better minimum. 5 bits is still small enough that randomly choosing a 5 bit factor will succeed 40% of the time. This is especially problematic since Shor's algorithm only has a 50% success probability per round, so you need some extra bits to be able to distinguish a correctly working quantum computer from a random number generator.