I was able to reproduce this bug using the old versions of piglit and mesa

  piglit:  1f39e07d44f7d78557d69c505168afb8eab913c6
  mesa:    c2a0600

After updating piglit to the latest revision, the test passes.  This
seems to be the correct behavior, based on Ian's comment.

Based on this information, I think it is safe to conclude that this
bug does not indicate a regression in mesa.  Instead, it is caused by
using a fixed version of mesa, but failing to pull in the
corresponding piglit fix.