Supposedly this should work for HTML content, but even the live examples don't show any difference between the rendering types. You can test it yourself if you screen cap the live example and paste it into an image editing program. Line up identical glyphs and use XOR blending on the top layer. The result will be a pure black image (ie, there is no difference between text rendered for speed, legibility or precision).
The only thing I have noticed is that sometimes rendering for 'speed' will increase the spacing between certain pairs glyphs.
Does anyone know how this is supposed to work?