True, PPLNS is ahead some of the time durring the first few days. But even before the bad luck streak, when everything falls below the PPS line, CPPSB is way ahead.
It would stay way below (at pure PPS value) during good luck times though, like all other *PPS methods. To me it seems as if it has less variance than other methods from that data.
I would really like to see another graph with equally "lucky" data or at least have this one marked with "unlucky simulation"...