With ECM, there is always a chance that a factor has been missed. "3155 curves" really means that if there is a factor of that size, then each curve has a probability of 1/3155 of finding it. The probability of not finding it is (3154/3155)^3155. It's well known limit that ((n-1)/n)^n approached 1/e. Hence the surprisingly large probabilty of about 37% that the number has been missed. However, it has a much higher probability of turning up with the next set of parameters, so in practice the missed numbers tend to show up quickly on the next level.

With ECM, you find the factor if a number near the factor is smooth enough - which number depends on which "sigma" is used for the starting point. So multiple factors can come out at one time, but it's more common for them to come out in different curves. With P-1, you will always get the product of all the factors that are sufficiently smooth. P+1 has this additional twist that depends (50/50, I think), on the starting point, so you might get none or one or a product on sufficiently smooth numbers in any one trial.