Maybe your numerical problems came from the parametric model you use for unsaturated flow. if you define some very sharp curves FEFFLOW may not produce meaningful results.
You have to keep in mind, that for the stability of the numerical simulation not the total value of capillar pressure or relative conductivity are important, but there deviation over saturation. So avoid any steep functions.
I made some good experience with the modified Van Genuchten approach. Here you have two possibilities to smooth your simulation:
1. Reduce the parameter A: this doews introduce some capillara difussivity into the simulation and helps a lot.
2. Use a value for the parameter delta that does not make the relative permeability curye to steep (e.g. 1.1 - 1.3)
Of course the parameters you use should have some physical meanings, but there maybe a large range in data.
The other thing you could try is a mesh refinement. Not so much in hirizontal direction, but in vertical direction. This shoud help much more the unsaturated processes take place in the vertical direction.
The vertical element length you should use has also something to do with the parametric model you use. You have to assure that an unsaturated profile could be computed an not just " wet or dry ". There is a rule of the thumb: if you use a (modified) Van Genuchten model the vertical spacing of your slices should be in the range of 1/A
Zebra