For a wavelength of 1.3 mm, we'd want the time tagging to be better than a quarter of the wavelength at least - say 0.3 mm. The speed of light is 300 mm/ns (a foot per nanosecond is the shorthand beloved of circuit and chip designers). So, for 0.3 mm, we're going to have to get down to a wavefront tagging accuracy of 0.001 ns.
No clock is going to get there, but if we can get ~close enough, we can use a procedure called fringe fitting to determine the clock corrections by looking at the wavefronts. (Does it line up this way? How about this way? How about now? Yes, it's as laborious as it sounds, but computers, eh.)
This is all in the calibration of data, before we do the Fourier inversion to create images - the magic of radio interferometry is that we can record the signal to disk while preserving phase. Optical photons can not be recorded and played back with phase preserved - optical interferometry has to split up the photon streams and send different parts to be correlated against streams from other telescopes, so you run out of signal quickly. Meanwhile, we can record radio waves at the 27 VLA dishes, say, and play them back for correlation on all 27*26/2 = 354 baselines, no problem. That's why radio VLBI is a thing, but not optical VLBI.
Even as a professional radio astronomer, the underlying physics is deep and almost magical.