In the 'ori' branch, I have update FieldVariable.get_global_node_tab() but noticed a bad convergence of examples/diffusion/subdomains.py. It was present even before the massive changes of the 'ori' branch - master (ec602a8162a96926afe09864adfe2e58a08c2503) version behaves in the same manner:
sfepy: nls: iter: 0, residual: 6.173330e+00 (rel: 1.000000e+00) sfepy: nls: iter: 1, residual: 2.606163e-03 (rel: 4.221649e-04)
If you put "'check' : 2" in the newton solver options, you will see that the tangent matrix is wrong. Can you check this? If it's a deeper problem, create a new issue, please.