You may want to take a look at the official NCEP/NCAR Reanalysis documentation. What I read in the data summary seems to imply that the monthly mean diagnostic surface temperature values are calculated from the " ... instantaneous values at the end of the 6-hr forecast."
-- Michael Bell