Ben,
   When you say transfer function you have be careful to identify the output. There are two common cases, S_phi and S_v. S_phi represents the noise in the phase.  S_v represents the noise in voltage. The relationship between flicker noise and S_phi is straightforward. The transfer function between the noise source and the phase noise response is inversely proportional to the frequency offset, and so increases at a 20 db/dec rate as the offset frequency goes to 0. The flicker noise source increases at a rate of 10 dB/dec as the frequency goes to 0, so the oscillator phase noise increases at a 30 dB/dec rate as the offset frequency goes to 0. It is important to realize that the perturbation of the phase due to low frequency noise can be unbounded.  

At higher offset frequencies, where the phase noise is small, the voltage noise tracks the phase noise. If you double the phase noise you double the amplitude noise. However that cannot occur when the phase noise is large.  While the phase shift due to noise can be unbounded, the voltage noise that results from the phase shift can be no larger than the amplitude of the signal itself. So as the offset frequency approaches 0, S_phi can go to infinity, but S_v is bounded by the signal amplitude and so must roll off.

The point I was trying to make in the JSSC paper is that if flicker noise is present, there is no explicit formula that describes exactly how and where the roll off occurs.

-Ken

Ben,
   I should point out that this low frequency roll-off in S_v generally occurs at such low frequencies that it is not usually of much concern. Also, you can estimate it by using the fact that the total area under S_v must equal the power of the signal.

-Ken