Heres my calculation, regarding "The Perfect N" in pplNs. CMIIW.

First, we calculate the time needed to solve N-shares

T= N x share-difficulties/pool hashrate

for example:
Tdbix1 = 3000 x 2GH / 500 GHps = 12 seconds
Tdbix3 = 100000 x 2GH / 15 GHps = 3.7 hours

Ideally, you would want to participate on every N rounds. Because  it doesn't matter how much work and how many shares you have solved, the only shares counted are the N numbers before a block found.

You can calculate maximum worst time for a block to be found with: Difficulties/Pool Hashrate
On dbix1 that is: 106716/571 = 190 seconds.
On dbix3 that is: 106716/25 = 4268 seconds = +- 1.1 hour

That's on CURRENT condition.


The randomness of block time makes it difficult when to hop.


Your average time to solve a share is: Share difficulties/your hashrate

If you have 100 MH/s, your average time to solve a share is 2G/100MHps = 20 seconds.

As you can see there is already a mismatch between your window to solve a share and your time to solve it,


So the mission is to find at least 1 share for every N shares submitted, that way, you will have a better chance to participate on every block, and not getting too many shares discarded.

We can calculate the perfect N
Perfect N = (Share Difficulties/Your hashrate) x (Pool Hashrate/Share Difficulties)
<=>
Perfect N = Pool Hashrate/Your Hashrate

So for a 100 MH/s on dbix1, the perfect N is 500GHps/100MHps = 5000

But since the pool hashrate always changes, finding the right number for everyone is impossible.


What about dbix3?

We can calculate the minimum hash rate to participate efficiently with:
Your hashrate = Pool Hashrate / N = 15000 MHps / 100000 = 0.15 MH/s

If dbix3 has the same hashrate as dbix1:
Your hashrate = 500000/100000 = 5 MH/s

as we can see, N = 100000 is probably too big

So my recommendation would be:
Either get more miners to dbix3, and/or reduce the N to the expected pool hashrate.