In my opinion, the gold standard dataset for global ocean surface temperatures is the UK dataset, HadSST3. A review of the uncertainties is given in this paper by John Kennedy http://www.metoffice.gov.uk/hadobs/hadsst3/uncertainty.html. Note, the UK group has dealt with the same issues raised by the NOAA team. I personally see no reason to the use the NOAA ERSST dataset, I do not see any evidence that the NOAA group has done anywhere near as careful a job as the UK group in processing the ocean temperatures.
I am also unconvinced by NOAA’s gap filling in the Arctic, and in my opinion this introduces substantial error into their analysis. I addressed the issue of gap filling in the Arctic in this recent publication: Curry JA, 2014: Climate science: Uncertain temperature trends. Nature Geoscience, 7, 83-84.