pe1chl I tested what hapens when I add the external reference and what when I remove it. The beacon shifts for about 230 kHz lower on the spectrum and stops drifting. The previous beacon disapears completely. For the test beacon I used a 24. harmonic signal from my Baofeng ham radio tuned to 432 MHz. There was only one beacon in both cases.
I also measure the signal on the crystal at both ends with an oscilloscope and compare it with the 10MHz from GPS in the XY-mode. I got nice and stable Lissajousse curve which means that the 25MHz signal was phase locked to the GPS.
@Mike5000 The signal level is about 3Vpp. I experimented with 1nF at first, but I got very big phase noise. I don't know why exactly, probably because the reference signal level was too high. Then I tried with several smaller capacitances and got much better results. I found the optimal value 15pF for my case.
In the picture above there are two added capacitors in serial, because I experimented with different values - just added lower capacitance to existed 1nF. In the final version I have only one capacitor 15pF (not shown on picture).