Both kinds of transformations are useful:
- real world transformations can be used to extract portions of the geometry, intersect or buffer it, offset and displace it in a way that makes the perceived offset zoom dependent
- pixel transformations are the ones that make the transformation zoom independent and can be used to move apart overlapping lines, dropping shadows and so on
The SE 1.1 specification seems to address, via a finite set of transformations, both of them, depending on the unit of measure used in the symbolizer.
This proposal is about going beyond, and having a pluggable extension point for people to create whatever transformation they need.