So here is what I see Capulo, this might actually be beneficial to add as a feature for the pool, yes, more work for me, but anyway here is what I see:

Out of the 50,000 or so solutions we have received over the last 10 minutes, about 8% of those are invalid.  Most of the errors (the lions share) is BLOCK_SOLUTION_STALE.  So that is actually pretty normal.  That happens when the block changes and a miner is still working on the solution and submits it late.

On Your miners however we have an 18.3% error rate (as compared to the standard 8.39% error rate).  The lions share of the errors are the same : block_solution_stale.  I see one miner in there a lot: capulo_wrk_21, maybe that particular node is hosed.

Anyway please check to see if the chain is synced, most of the errors are on the old version too, I do not see a higher incidence of errors on the new version.  You might have nodes behind an IP on a fork together.

I will go ahead and see if I can add a feature to the pool to convey the error ratio of the user up into either the leaderboard, I guess leaderboard might be OK then we can all see who is hosing the pool the most descending, it would be a nice feature.

Im going to work on the yobit paperwork first though!