After the previous reply, I asked on the mailer for Modeller9v8 about refining Modeller-built structures. Two members replied they’d had good luck with Rosetta relax on proteins even larger than mine. So I tried it with relax:fast. I started the job on a 2.33GHz linux box just after lunch on a Friday and it finished sometime Saturday night while I was asleep. The output has a very good MolProbity score – 1.93. I’m happy.