I'd really like to know how the technology works. Blog entries and web pages welcome, but the ideal answer will include one or more references to published papers in the professional literature.
bottom line: do not try to render more pixels than what is possible by the screen. you do not need to touch data which is not relevant for the final image because it can not be seen anyway. lets take a normal image viewer:
so, normally you create a surface you can draw to and then you iterate over the original data and paint the pixels on the surface accordingly. if you "zoom closer", you draw 4, 8, 16 pixels per zoom instead of the original. and if you "zoom away", you draw only every 4th, 8th, 16th pixel (or, to improve quality you calculate a new color based upon the pixels around the one you would pick).
the problem with zooming closer: you tend to let the window system clip away pixels outside the viewable area, though you still have a, lets say, 8k * 8k big surface. your screen is 1980x1020 .. so thats a huge waste of memory (and speed in the end). it would be better to find out what is visible BEFORE you start to get data from the original source.
the problem with zooming out: you have to calculate over and over the final pixel. it would be better to have the original data as best as possible AND at some lower resolutions.
so, seadragon combines several things: calculating what the user can see, accessing the data to fill only the viewable area, and do that filling very very fast. additionally be prepared to fetch data that is "close" to where you look at right now. it is uncommon to zoom into a picture VERY close and then jump INSTANTLY to the other corner of that image. normally you pan around in one region a little bit so one could prefetch the data as well.
better and deeper explanation: