Have you considered taking a fee for the pool? Maybe something like f=c=0.005 (1% average). I don't know if many people would mind, and it might make it worth your while even without participating yourself.
Meni, would you care to comment on Eligius' new MaxPPS-system?
[Edit: The following should be correct for MPPS.]
It isn't any good if you ask me. Sure, if you make an oath to mine for eligius forever until the end of time, it will even out eventually. But if you just want to mine for it for any limited amount of time, the odds are against you.
If you mine for such a pool for the duration it takes the pool to find on average 1 block, you get a penalty of 1/e on average, so your expected payout is only 63.2% of the average in other methods. If you mine for the duration of finding on average L blocks (where L is an integer), your penalty proportion is exp(-L)L^L/L!, which by Stirling's law is roughly 1/sqrt(2 Pi L). So for L=100 your average penalty is 4%.
Given that pools are supposed to drastically decrease your variance while having insignificant impact on your expectation, I find this reduction in expectation, or alternatively the long-term commitment, unacceptable.
Edit: Actually, I'm still not sure how exactly MaxPPS works. Whatever it is, though, I suspect these calculations are at least approximately true.
[Edit: The following is for SMPPS]
Reading the more detailed description it looks like it doesn't work like I at first thought. It looks very difficult to analyze, which is why every time I try to think about it I come up with different conclusions. However, according to my current understanding it is doomed to failure. What will happen is this:
The pool balance (total earned - total owed) follows Brownian Motion.
Which means that it will reach any given level with probability 1.
So at some point the balance will be deeply negative.
So any newly submitted share will receive only a fraction of the expected payout.
Seeing this, miners will leave to greener pastures.
This will slow down the recovery, until everyone is fed up with it and the pool collapses.
Everyone who still has a pending reward will never receive it.
Note that there are several perfectly valid hopping-proof methods. PPLNS is hopping-proof and has less variance than the geometric method (mine), but it involves the extra complexity of crossing round boundaries. It has two variants, with or without double-counting shares.