[snip]
Mike, thank you. I just had one of those "Doh! Why didn't I think of that?" moments. I think you may have solved the mystery. The loop bandwidth is only slightly wider than 1 KHz, so of course there is very little attenuation of VCO noise at that frequency! I've just checked the predicted response of my loop for noise injected by the VCO, using my SCILAB model, and it agrees to within a few dB with the measurements. I'd like it to be closer though .....