Some amount of relative drift in inevitable when the sources are independent, but yeah this suggests that the precision is only on the order of 1/48k. I cant imagine the hardware oscillator being that imprecise; crystals and mems are both better than 1/1M. I wonder if it's somehow software related. Ive seen some mysterious stuff with built in audio like variable input latency. if the output latency fluctuated similarly it might effectively introduce a low frequency modulation.