We have a web application which shows the (United Kingdom) Census data on thematic maps using client-side code (based on OpenLayers) and KML.
// This section can be skipped
At the moment it works in the following way
We store area shapes as KML in database (just shapes, no styles, colours or other data)
We also store the Census data against each area (e.g. area="Output area E01000123", time period=2011, variable= "% of single people", metric value = 34.12) in the same DB. Of course, data structure is normalized.
A user selects a region (e.g. London) and a variable
We get all the small areas within the selected region, generate the KML and apply colors based on metric values.
This approach works well when the number of small areas in a region is not too large. In some cases the number of small areas might reach 5000. In this case client-side code is too slow in IE8 (which we have to support). So I guess we have to generate an image from the shapes server-side.
What is the most appropriate technology?
In the past we used GeoServer with PostGis backend. It was not too fast and quite unstable. I can explain how we used it.
I decided to use GeoServer + PostGis. GeoServer serves images. When a user selects an area, the client-side code creates a custom SLD and passes it to GeoServer. When no or few areas are selected, it is very fast (~1sec), even with 30000 polygons in the layer. If a user selects e.g. 2000 areas, it takes several seconds (I guess because GeoServer has to process a huge SLD file).