Just to let you know how it panned out …

To cut a longish story short, it looks like LatheJack's suggestion that the D1-4 cone was binding before the mating surface on the backplate hit the spindle flange was right. I clocked the mating (front) surface on the backplate for perpendicularity to the spindle axis and there was anything up to 0.4mm 'wobble' – not repeatable at the same orientation of backplate wrt spindle axis. I noticed that there was a thin circular witness mark where the nose of the spindle cone had dug into the receiving cone on the backplate. So I polished off the witness mark, remounted, repolished etc etc. Eventually got it down to 0.04mm repeatable wobble, with a broader witness mark. At that point I decided to call it a day (I was getting a bit bored!), so reassembled and remounted chuck. Now about 0.05mm runout on the test bar, still not wonderful but the chuck is useable and more importantly I understand what's going on. I'm sure I can improve things further by taking a skim off the backplate now the error is repeatable.

I think the reason it seemed OK before I meddled was that the cams had been tightened with unreasonable force when the thing was assembled – there is some mangling of the pins on the backplate to support that idea.

Thanks to all for your replies, Rob.