I think it needs to be multimodal: a sense of magnification, as here, but easily traversing the intended visual flow of the page elements, and also keeping some sense of overall context regardless of magnification level.

One big goal is to be able to present 19th century newspapers online, without needing to break them into individual columns at all scales.

But it could get tedious. Look at this book, for example—we have five 3000-page volumes of this junk sitting here waiting for a solution. Scanning it without snipping off page ends was too challenging for our Google buddies.

Let alone the ridiculous amount of effort involved in slicing and presenting a few hundred thousand page scans as individual multiscale images… let alone reading them.

Still thinking.