For raster data you run out of memory quickly; On uDig trunk I have worked
with an "image moasic" (think a shapefile where each row indicates a raster
file on disk) of several thousand MRSID files; each file was 20 GIGs...
there are a couple competing projects to use PostGIS as a spatial index for
directories of raster files as we speak.
Even with just vector data we manage sneak up to the limits of memory
fairly quickly - it is one of the reasons I enjoy GIS.