You're probably seeing Mach banding.  

This is a fundamental limitation of Gouraud shading which just
linearly interpolates color intensities across the triangle
vertices.  The human visual system is very good at detecting
2nd-order discontinuities across shading gradients, which will
occur at the boundaries of your triangles with Gouraud shading.

With directional lighting the light source is at infinity, so all
the normals of your flat surface are at the same angle to the
light source and will be lit the same.  With point and spot
sources each normal will be at a slightly different angle with
respect to the positional light source and will receive a
different shade value when lit.  These differences are made much
more apparent with flat shading.

Ideally you would use Phong shading, which interpolates dot
products for each pixel across a triangle instead of just color
intensities.  This is available with modern pixel shader GPUs.  I
think Java 3D 1.4 supports that stuff.

You can simulate Phong shading to an arbitrary degree by dividing
your surface up into smaller triangles.  When each triangle is
about the size of a pixel you'll have perfect shading for
positional light sources, but you'll of course run slower and use
lots more memory.

You might also be able to minimize the mach banding with Gouraud
shading by turning down or completely eliminating the specular
coefficients of the surface.  Specular lighting involves
exponentiation of the dot products at the triangle vertices and
will exaggerate the tesselated nature of the surface.

The easiest workaround is to just use directional lighting.  It's
easier for the GPU to compute as well and you'll have better
performance.

-- Mark Hood