It makes assumptions about text positioning that are
not holding with subpixel positioning. There is no
guarantee that the next word in a multi-word text
starts on an even pixel boundary, as it does when
you break the text into multiple, separately rendered
blocks.