very interesting from a scientific point of view. Is there visible practical use during 'normal', non-Matlab/camera assisted lab experiment environment ?

I guess after going through all calibration steps, either on the PC or on -possibly- a N900 version of the calibration tools, using both distortion values into the Fcamera driver might possibly result in better photos. But would the results be visibly clearly better ? Any idea ? Any examples to show made by similar hardware ?

After all we're not talking about the latest 21 Mpixel DSL Canon EON-1Ds beast, this is a 'humble' N900.

Also, are the production parameters of the N900 (or in general) camera module lenses tight enough to do the calibration once and then use it for all N900s ? Or are there normally significant differences between production runs possible ?

Good luck with the calibration anyway, it sure looks you know what you're doing !!

TO come back to your question : Look here
http://wiki.maemo.org/N900_Hardware_VGA_Camara_module

And click through to the datasheet :

Pixel size 2.2 μm x 2.2 μm

There you go Let us know how things are evolving ...