The 8×8 regions have been abandoned because even when they were just bare ground they took far too long to render. This made border crossings a nightmare - even though you rarely came across one.
Instead, I have re-factored the landscape to be a grid of 4x4x4x4 - that's 16 regions in a 4×4 grid, each region being a 4×4 region (taking up 16 grid coordinates each). This gives a landmass covering the same area as 256 standard regions. This makes it the same size as when it was 4x8x8. However the landscape renders much faster.
That just leaves the small problem of the heightmaps.
In a 4×4 region, it is relatively easy to create a heightmap that covers the entire region. If you had 16 1×1 regions (which is the same as a 4×4 when arranged in a square), you could take the same heightmap and cut it into 16 pieces and load them in the correct order into each of the regions. This would give you a seamless landscape.
My current arrangement of 16x4x4 regions allows for a much more varied landscape, but the usual tools for creating these landscapes, such as L3DT, don't allow for such a large area.
I have to rethink this: I can cut the map easily enough in GiMP, which provides tools to do exactly that. Therefore, I just have to create the original map - and that's where the problem lies.
Ideally I want to start with a crater formation, so the outer edge will have at minimum low lying hills (to hide the default landscape of what lies beyond a region boundary). The very edge though, should be at zero height, to avoid nasty artifacts (such as being able to see beneath hills).
I want to be able to define a structure within the map too, so I could say for example 'there is a high ridge here' or 'there is a mountain here'.
It seems then, that I start with a template image, and generate a map from that. The map will have to ensure smooth transitions between low- and high-points, so if I want cliffs later I'll probably need to edit, or I could define a cliff within the template.
The template would use various colours to determine how the map generator should understand them (e.g. one colour for high points, another for cliffs, another for low points etc.).
The generator then could be tweaked and tweaked until the output is satisfactory. I could then get it to read in the previously generated output and slice it into heightmaps the correct size for each region, or do it manually in GiMP.