I'm not entirely sure I understand your point. All of these examples were likely designed in PhotoShop (or Gimp) and then realised as PNGs, GIFs or in this case CSS. Beyond that adding a button is pretty simple regardless how you do it; however, how that button is rendered, which is the point, is certainly simpler with an image (and can be made trivial for developers with appropriate use of CSS).
I think what tptacek is getting at is programatically describing the button which then gets created in an automated way rather than manually creating it using a graphical tool such as photoshop.
I think it is highly practical, since one very skilled designer and CSS programmer with enough programming skill (or a team) can create a template that many other less skilled and less experienced designers can then take and use at will. The template, while being rather complex compared to the hand-coded and "beautiful" markup that Thomas was talking about, can also have enough dumb logic in itself to degrade gracefully on legacy browsers.
Wouldn't the degraded version depend upon the same old Photoshop approach? If so, you might as well keep using that approach until all major browsers support the necessary components of CSS3.