I took some time to look after this bug. The problem seems to come from the sample_cube function in sp_tex_sample.c, when the face of the cube map is determined using the four pixel's texcoords average.
At the edges of the current rasterized primitive, some pixels may not be part of the primitive (ie. masked), leading to an incorrect average and therefore an incorrect face selection.
-- GitLab Migration Automatic Message --
This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.
You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/207.