gl_Normalscale is supposed to be 1/sqrt(m_31^2+m_32^2+m_33^2) with m_ij
being the components of the inverse modelview matrix.
After a call to glScale(1, 1, 2) the _inverse_ modelview matrix is a
diagnoal matrix with the elements 1, 1, 0.5, 1. gl_Normalscale should thus
be 2 rather than 0.5 as presumed in the test.

Because of color clamping alter the glScale call rather than the expected
result.