Not always. In my experience there are two things that tend to break when you try to render to canvas and apply the empirically measured dimensions to HTML text: (1) subpixel rendering because it does still exist despite high DPI displays, and (2) kerning and hinting don't always give you the same metrics inside/outside canvas, especially for non-Latin characters.
