I think we completely misunderstand each other.
I am talking about:
my FPGA hardware doing 0.01% rejects or less on Bitminter using Stratum. Same hardware, exactly same configuration doing 15% rejects on p2pool. I tried to tweak config and got slightly better results about 10% but that's still not the main case. 15% - That's reported by miner and by p2pool stats as well. That's wasting resources, nothing else. When we invest our money and time in something, we should do it right. FPGAs just react too slow to change nonce every couple of seconds, to top of that add remote node's bitcoind latency + network latency + time to prepare getwork etc... This is just not working for LP= 10s, it's a disaster. No one is mining of remote p2pool nodes as it will just not work as it is right now.
If you guys are saying there is no solution for that - so beware, because one day p2pool will die.