My (simple) noise model of the AD5066 is an 8k resistor's thermal noise plus the reference buffer's thermal noise (~60nV/rt-hz ??) multiplied by code/(full-scale code).Does that sound reasonable?
That is a reasonable approximation. The reference buffer noise may be a bit better. The plot below is from the AD5062 which uses a very similar buffer to the AD5066. A value of 25-30nV/rtHz may be more accurate but 60 is a good value to keep some safety margin.