When using Verilog-A models with SpectreRF there are two things to worry about. First is hidden state, which is discussed in the paper that can be found at
http://www.designers-guide.com/Analysis/. If there is hidden state, then the RF analyses generally just refuse to run, or if they do run, produce convergence problems. If your models are running without difficulty, then hidden state is generally not a problem.
The other problem that might occur is present only with small signal analyses, such as AC, noise, PAC, PNoise, etc. It occurs if the model contains an abrupt discontinuity, such as might be the case if one were modeling an ideal comparator. This is a conceptual problem more than anything else. Consider a model that produces 1 if the input is positive, and -1 if the input is negative. A small signal analysis will indicate that none of the small input signal will pass through the comparator (its gain is zero) because the input signal, being small, is too small (by definition) to trigger the comparator. Thus, no noise (or jitter) will pass through the comparator. This is not really correct because 1) no comparator is ideal, and 2) when it comes to comparators, no signals are small if the input is biased at the threshold comparator. If your model does not contain abrupt discontinuities (the actual function is discontinuous, not just the derivatives), then this one should not be a problem.
If you'd like to know for sure, post your model and I'll try to take a look at it.
-Ken