Possibly there are many autofluorescent cells that are GFP neg among the GFP positive sorted cells. In the histogram it is often difficult to see or gate the difference between autofluorescence and positive cells. Autofluorescent cells can be better seen as a diagonal population in a DotPlot FITC (530nm) versus PE (580nm). In contrast to the autofluorescent cells, GFP positive cells have a stronger fluorescence in the Fitc channel and can therefore be easily separated from the autofluorescent cells (without compensation) which can improve the sort purity.

thanks for the advice, I'll try again. but it's curious since I've done gate using dots plot and sorted mid-bright population and respect to control (auto) there were about 2 log decades of gap between negative and positive (ie P3 gate), so it's strange to see so many negatives in post sorting.

I second Toralf on the histogram issue. In addition, the GFP signal is quickly lost when cells are dying. Your post-sort GFPneg cells show a slightly lower FSC profile than the positive ones. I recommend using a viability dye to see if the GFPneg are dead/apoptotic. If they are, they might have been GFPpos at the timepoint of sorting, but then suffered