The attached image shows a slice from the G-code viewer,
Gcode Analyzer.
The hemispherical thin wall (0.04" /1.02mm) part was sliced by Slic3r 0.9.8. While the erratic nozzle path is troubling, the bigger problem is the infill gaps. The slicer plots a single inner and outer perimeter, but a broken infill.
As the slices approach the apex, each layer becomes progressively wider as it approaches the tangent of the sphere. These layers infill with a rectilinear pattern, regardless of the infill pattern selected, but they also still have infill gaps eventhough there is sufficient area for a proper fill. (See GCode Analyzer 4.jpg -- This image shows support material pattern in the middle of the circular walls.)
Setting Fill Density to 1 does not fix it.
Setting Solid Infill Every 1 layers does not affect the output.
Changing fill pattern to concentric does not change the infill pattern at all.
Minimum perimeter is set to 2, but only one perimeter is every made.
Changing the STL normal deviation to 5 degrees, did seem to have some afffect, but mostly just resulted in a bigger g-code file.
It would be nice get consistent infill around the perimeter. I bumped the wall thickness to 2.15 mm as a test, but ended up with even larger infill gaps, see GCode Analyzer 5.jpg.
Eugene