Kinda a catch 22 there... More hash rate == Quicker Block solves == lower variance. (PPLNS payout) p2pool has less total hash rate so therefore it will have higher variance which will lead to fewer people using it which leads to lower total hash rate....
That's not so bad. People leaving are those with the worst configurations (network issues or underpowered node), who don't have the sysadmin skills to maintain a node or who can't stand the variance (due to very low hashrates or luck misunderstanding).
Regular users of p2pool are those who know how to set it up, maintain and optimize it: they build a high quality network of nodes that should outperform any centralized pool. With lots of well configured nodes broadcasting our blocks, we should have lower orphan rate. As anyone verified/quantified this by the way?
Last time I studied the orphan rate for solo miners on alt-coins, I found a block collision rate that was equivalent to a 3 to 5 seconds window where a block would be orphaned. This should amount to a ~0.5% orphan rate on the bitcoin P2P network. This is over-simplified by assuming most pools use a single bitcoind to broadcast their block (some have severals or have direct connections to other "friend" pools).
Is there some archive of P2Pool found blocks somewhere? Looking at the last thousand and comparing to the largest pools we should have an idea of our performance.