Many VCOs are not really operating small-signal linear
(e.g. CMOS current starved ring oscillators) and many
supply-noise challenges are not really small signal either.
AC analysis depends on all of the blocks' DC solution
having relevance / decent frequency domain modeling.
It's an OK place to start.
But I think if I cared a lot about things like switching
power supply induced spurs, I'd be trying to model the
expected real waveform of the supply and run transient
sims with that at 1X, 0X and 2X amplitude on the "noise"
(ripple and ringing, most likely - not random) and FFT
the output waveform, just like you'll someday do on the
bench. Yes, this will take a bugger long time to execute.
But it's liable to show you things small signal analysis
won't.