Great feedback!
I'm not sure the concept of a secured wallet that needs to maintain thresholds for users and process payouts dependent upon when a user crosses that threshold can be translated into a decentralized model. Centralized pools like GHash/BTCGuild/Eligius/Slush all have their own payout models and can pull this off because they are indeed centralized.
You may be right, I tweeted a few of the Dark Wallet devs, some of which I have spoken with before, maybe one of them can share whether they think it is technically possible or not.
Maybe we do nothing at all, and just accept the fact that unless you increase your own hashing power, you will experience more and more variance as your hash rate becomes a smaller and smaller portion of the p2pool network's rate.
This is pretty much where we stand now. When/if a multiple PH/s miner hits the pool the minimum hash rate to even hope for a regular payout will sky rocket, and p2pool is a very attractive option if your building out a large mining operation.
Even if we go with a multi-tiered approach, as I suggested, how do we define those tiers? What becomes the "cutoff" value to move from tier 1 to tier 2 to tier n? How many of these tiers would be "enough" to effectively reduce the variance?
I like the concept, but it will require huge adoption to work. Correct me if I'm wrong, you would effectively be diluting any variance reduction we gain by growing in pool hash power by the number of tiers that were offered?