No, we're talking about expected reward given to you by someone else's share. This is a p2pool conversation, not a traditional pool conversation. You have a probability of other people rewarding you in the shares they find after you find your own share. If there are more peer shares that are supposed to reward you, then the size of the reward that each share would give you (should that share also be a block) will be less. The point I have been trying to make is that the expected reward you get for mining a 1-diff share is not dependent on the number of peer shares that your reward is distributed over.

My simulation for the naive PPLNS-on-DAG system gave about 1% typical variation in payouts when the number of shares per level of the DAG varied uniformly between 1 and 10. Note that the variation is not predictable: you can't guess how many shares per level will be mined in the future unless you're actively manipulating it (and large enough to do so). For the PPLKD system, your reward will be split over a constant amount of other people's work, so the variation due to changes in M should be basically 0%.