I again want to highlight the importance of dimensions that are not scalable.
If you remember Voronyj diagrams[1] that Natalia did, usually you do not care about the size of borders, or the size of sites compared to the rest of diagram. But as total size of a resulting diagram is not predictable, and you scale it to fill your viewport, you end up with a different size of borders which is sometimes inappropriate.
Again, when you have a scatterplot. Sometimes you have clusters, and sometimes you have sparse plots. I think that it can be essential to be able to say: “I need my elements to be always 5px in radius”. This way you can always see them when you zoom out and they are not overlapping when you zoom in.
I understand that this may be hard to implement. Just wanted to highlight because maybe someone is planing to do something cool on the next weekend :)
Uko