I really like your description of different device classes, but I think it’s important to not only classify them by screen space and connection speed and bandwidth costs/constraints, but also by the applied usage paradigm. For example, on a PDA, we generally don’t have a mouse. Thusly, there is no constantly moving cursor on the screen. Instead, we only have discrete “click” events. But this, in turn, renders “rollover” events non-functioning on such devices because there is no such thing as a “rollover” when we only have discrete clicks. However, it’s easy to dig up some websites that make extensive use of “rollover” events, such as when expanding menus when the mouse rolls over a headline etc… All this does not work (or would feel awkward, since we don’t want to drag our finger or stylus over the screen like a mouse) on a handheld.

Instead, the more diverse web-browsing devices become, the more we have to employ what you refer to as ‘Multi-Web Practice’, in order to create website front-ends that blend in well with the usage paradigm of each device, as well as its physical (and possibly economic) properties.