There are 2 differences between what happens in the itemeditor and the event_filter example.

a) In the itemeditor example the content of the plot remains unchanged. So when the item is dragged the content can be restored from a pixmap cache without having to do a replot.
b) Depending on the Mask parameter, the pixels written to the graphics stack might be limited to the pixels, that are really changing.

As the event_filter has almost no content, the effect of a) should be no factor - for heavy plots it would be significant. So in your case I would expect, that b) is what you are suffering from.

In case of remote desktops it is obvious, that pushing a fullscreen image ( width * height * 4 ) over the network is slower than a smaller one. Here it might be worth to go back to Qt4 ( graphic system native ), what is much faster for vector graphics. But as you are not using a remote desktop it means, that flushing a RGB buffer is a horrible bottleneck - something that will hit you in many other situations as well.

Uwe