Two days ago, while trying to recover the PLL lock, we found that the spectrum analyzer we used to montor the beating note (HP 8563E - see pic1) was not working properly.

It shows a reduced bandwith: the beating note disappears when its frequency becomes larger than about 130 MHz and the peak amplitude was lower of about 15 dB.

We replaced it with another one (HP E4411B - see pic 2) which seems to work better (correct bandwith and peak height). 

The behaviour seems to suggest a impedance matching problem but since the its settings and the configuration of the setup are exactly the same used before we cannot understand the origin of the problem.

More investiagations need to be done.