Subpixel rendering is, of course, very helpful. It's really preposterous though that we have to resort to abusing fonts to get subpixel when implementing it for arbitrary vector rendering is so straightforward.
The other big problem with using fonts for hinting support is that people have differing opinions on the value of hinting for fonts, but everyone would probably agree on the value of hinting for icons. If you disable hinting because it hurts your eyes (or because you use OS X and don't have a choice), then you're going to miss out on icon hinting as well.
Seems to me what we really need is a client side solution for rendering arbitrary SVG with either manual hinting or icon-optimized autohinting (which might be too general of a concept to be doable).