This was working very well a short while ago, but seems to have regressed significantly. Rubberbanding works fine as long as no selection is involved, and becomes extremely sluggish, spiking the cpu usage as soon as items are selected, even if the selected items don't change.

So when I ran into this my laptop was running off of battery. When plugged in, it's not nearly as bad, though 2x-3x more cpu usage as soon as any item is in the rubberband. So, this isn't really a regression but just points to some room for more improvements. Maybe, while rubberbanding, we should cache the rendered items in a bitmap, which would allow us to more efficiently blit them and render the rubberband on top.