Thank you for looking into this ChrisM and Eirikur. My last contribution, i.e. the bit about special case data versus random data in the #hidvals table led to the further (unasked) question re why the #hidvals statistics for special case data (a single value for each of the fields start and finish) were apparently being ignored when a UNIQUE spec of some kind was defined for the temp table.
It is heartening to hear you have replicated the case. I am looking forward to hearing your conclusions.