So, while wondering about shape simplification in the eternal struggle for optimization, I find out (here on SE) that Geoserver and PostGIS can and will (by default) perform the trick on the fly (see Small features don't draw when zoomed out using Geoserver WMS), all by themselves.
A number of questions come to mind (does it work with shapefiles too ? at which point of the process does it happen ? obviously the renderer must simplify the shape in any case, simply because of available pixels - but does a non-simplified shape that gets simplified on rendering have the same weight resource-wise as one that's pre-simplified ?), but to pick one:
how much of a resource hit is it, compared to preparing simplified versions of the geometries beforehand ?
The spider sense says it should be significant, but Geoserver has surprised me several times already in the brief time we have known each other.