Bisect shows:bce58e155db7202a98642c10e6132dee4e08162b is the first bad commit.
commit bce58e155db7202a98642c10e6132dee4e08162b
Author:     Eric Anholt <eric@anholt.net>
AuthorDate: Wed Jul 4 10:52:32 2012 -0700
Commit:     Chad Versace <chad.versace@linux.intel.com>
CommitDate: Tue Jul 17 08:17:56 2012 -0700

    intel: Convert to using private depth/stencil buffers (v2)

    This means that GLX buffer sharing of these no longer works.  On the
    other hand, just *look* at this code reduction.

    v2:
      - [chad] Fix intelCreateBuffer for gen < 6. When the branch for
        !screen->hw_has_separate_stencil was taken,
        intel_create_private_renderbuffer was incorrectly not used.

      - [chad] Remove all code in intel_process_dri2_buffer for processing
        depth, stencil, and hiz buffers. That code is now dead.

    CC: Eric Anholt <eric@anholt.net>
    Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
    Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

Did you change only mesa and not the kernel between the working configs and the failing configs when you started bisecting?  Was the failure 100% reproducible in the bad commits and 100% not reproducible in the good commits?

It always pass on commit 433ff3e16e8e090fd3a1bf427e61f3e5971a5740.
It always fail on commit bce58e155db7202a98642c10e6132dee4e08162b


OpenGL Report.
    Vendor - 'Intel Open Source Technology Center'
    Renderer - 'Mesa DRI Intel(R) Sandybridge Desktop x86/MMX/SSE2'
    Version - '3.0 Mesa 8.1-devel (git-bce58e1)' (3.0)
    GLSL Version - '1.30'
    Context Flags - None

>> Occlusion Query (occlusion_query)  test:
--> 2.6 - advanced.fbo subcase:
    File - /GFX/build/testsuite/Oglconform_31/oglconform_31/src/OGLconform/occlusion_query.c, line - 560.
        Occluder is less then occludee therefore passingSamples should report
        500. Instead passingSamples = 0
        OcclusionTest test case 0 failed


TEST PURPOSE:
Verify that occlusion queries are possible in FBO.

TEST METHOD:
Test creates FBO and runs scenario from testcase 2.1 in it (3 queries, one passes, two fail).

PASS CRITERIA:
Passing query retuen 500 pixels passed, failing return 0 passed.
--------------------------

--< 2.6 - advanced.fbo subcase failed.
<< Occlusion Query (occlusion_query)  test failed. (1 of 1 subcases)

Intel Conformance failed.
Total Passed : 0
Total Failed : 1
Total Not run: 0

Run: ./glxgears on Huronriver with mesa master branch, system hangs.

Bisect show:bce58e155db7202a98642c10e6132dee4e08162b is the first bad commit.

This is fixed by Eric's kernel patch titled:
"drm/i915: Don't forget to apply SNB PIPE_CONTROL GTT workaround."

It's available on the mailing list, but isn't merged yet:
http://lists.freedesktop.org/archives/intel-gfx/2012-July/019344.html

Reassigning to Daniel; he can close this after he merges the patch.

Should be fixed on -fixes with:

commit e844b990b1df9242bb91b7d490552f3198946838                                                                                      
Author: Eric Anholt <eric@anholt.net>                                                                                                
Date:   Tue Jul 31 15:35:01 2012 -0700                                                                                               
                                                                                                                                     
    drm/i915: Don't forget to apply SNB PIPE_CONTROL GTT workaround.

Can't find out "drm/i915: Don't forget to apply SNB PIPE_CONTROL GTT
workaround." on  -queued kernel commit:ab3951eb74e7.

Verified. Fixed on Mesa master a3d9d7ec79d6f7205fab2324e47d8ea185431de0 and
-queued Kernel cd7988eea561a70a4f98e431c1395f913672d626