Zero copy nan handling functions, faster mean() and std().
ENH: Add generalized ufuncs nansum(), nancumsum(), nanmean(), nanstd()
If this is not significantly faster then Bottleneck (http://berkeleyanalytics.com/bottleneck/intro.html), I would say that this should not be added. Even if it is, I'm not sure that the gain outweighs the decrease in maintainability.
By faster mean() and std() I'm not sure you are talking about the nan versions, since this does not make any changes to the current implementation. Improvements to current mean() and std() ufuncs should be a separate issue.
Time to close this.