But, I think your math's result is mainly due to variance recently introduced by the recent leaps in network difficulty and the increase in the pool's overall hash rate over the past year, both add substantial variance to your broad all time difficulty_sum/accepted_sum stat. Those sums include 19 months of counts, but, over 50% of the shares accepted over all time by the pool have been accepted in just the past 40 days! 40 days is not enough time for variance to level out.
I apologise in advance for sticking my nose in, but a reliable way to estimate long term luck is mean(diff 1 equivalent shares per round/network difficulty at which block was solved) and then calculating the Erlang distribution CDF. Over the last 2123 rounds, the mean shares per round /D is high at 1.057, and the CDF is 0.977 - so pretty unlucky. However for the last 1000 rounds the mean shares per round / D is only 1.023 with CDF 0.769 - not very unlikely (or unlucky). I also took a quick look at some boxplots and they didn't appear to be unusual.
Without completing a proper analysis I can't be sure, but Eligius' luck doesn't look unlikely.