The failure to replicate was of their test, not of the initial observation. Specifically it was considered interesting why the distribution of grades in CS (apparently typically two-humped) was different from eg mathematics (apparently typically one-humped). As far as I know this still remains to be explained.