Thank you for the question. The FDTD algorithm makes few assumptions beyond discretizing time and space, so it is, in general, fairly accurate. Theoretical models usually make several assumptions and approximations in order to obtain analytic solutions so they tend to be less accurate than a full 3D FDTD simulation.
The varFDTD solver essentially makes an effective index approximation in the z direction. This is accurate for planar structures where there is negligible coupling between slab modes, for example in SOI structures. The varFDTD solver is mostly intended to be used for quick initial simulations to narrow down your range of parameters. Generally, if you are simulating a planar structure, you should perform your initial parameter sweeps and optimizations with varFDTD to save time, then perform your final simulations with the full 3D FDTD solver for maximum accuracy.
Calculating the coupled power into a ring resonator is a common application for the FDTD method. It should be more accurate than any theoretical model that I am aware of for couplers. You might be interested in these posts that discuss extracting the coupling coefficient from FDTD simulations:
In summary, the varFDTD simulation should be fairly accurate for this application, but you should perform your final simulations with the 3D FDTD solver to get the most accurate results.
I hope that helps. Let me know if you have any questions.