There appears to be a clipping regression for g965.
in addition to the test in the summary, spec.glsl-1_10.execution.clipping.clip-plane-transformation pos_clipvert fails in the same way.
Both tests pass when run independently. Looping several thousand times on the single test does not reproduce the failure.
They fail when piglit is running the whole suite, either in parallel or sequentially.
Also, you can reproduce the failure by looping on the test and running some other workload like shaderdb.