I'm guessing it's just a matter of the equation denormaling it all - 'c' in that equation is the output of the formula itself, so as Songtime increases, it drags through states where the output (three bands in a parametric eq, the first subharmonic in one of the synths, the X parameter of a tweaked Melodic Bandpass Love Philter preset, not to mention all the peak controllers on associated channels being pushed around from the output of these and all of their linked parameters) reaches some sort of unity/feedback state. I think I ended up fixing it by rendering at a slightly lower PPQ. Regardless, it's out of FL sounding the way that I want - with maybe slightly more jagged modulation, but it's a lot easier than the hell I'd have to go through in putting all of this into a Patcher and going full-on debug mode with the formula.