I didn't measure the frequency oscillation width, but it was clearly visible at every silence. I suspected a sampling artifact, maybe in conjunction with an output subtone, because I had never seen the same effect on other NBFM carriers. As a cross-check I tuned another UHF carrier, which looked clean:
This unmodulated carrier, receiver over the air (see the signal disappear in the waterfall as I moved around the antenna), was clear and stable, which excluded an instability within the RTLSDR hardware.
Few days later I met the repeater technician (on the repeater itself) and he confirmed the PLL instability in the Motorola transmitter. So, it is not a sampling artifact but I was really looking at an unstable signal.
This oscillation is about 1 Hz, which is hard to spot on the equipment I have access to. But with the RTLSDR + waterfall software it was easy to see. Not bad for a 20 USD device that covers between 60 and ~1400 MHz. Good to know for my 70 MHz building activities.
