No, that all looks very good.Your original question was about the inconsistency between those two sets of measured time intervals, and I don’t think that the timing stability of your display is really relevant to that.
But regardless, the fact you were seeing that level of variability in the first set of numbers was simply due to the structure of your code. If you time your RT correctly from after the first win.flip() of your stimulus, you should be good.
Why that variability was markedly reduced when running full-screen remains to be explained.