It's not a reactance (impedance varies with frequency), but something second order (a resistance that varies with power and frequency), so you'd have to draw up a circuit first to solve for that.
Thermal diffusion processes are lossy, i.e. first order or fractional order frequency responses -- so you'll need to build some kind of active oscillator with a companion time constant (another phenomenon with an inverse frequency response, or use an impedance converter to synthesize one, or a regular capacitor/inductor).
For example, if you bias up an incandescent lamp beside a photodiode, and step-change its voltage, it will respond gradually, and with enough gain in a feedback loop around there, you'll get an oscillator. It's not parametric in the way you envisioned, I think, because you're measuring the temperature directly (with a light detector), not through its electrical property alone.
So further to that, you could replace the photodiode with an RF bias tee, measuring the incremental resistance; an LO and mixer provide excitation current and voltage detection, so that you read out resistance as voltage. (This can be as simple as a clock oscillator and quad analog switch, and some LC networks and maybe transformers for the coupling. And if it's done at a more modest frequency, which it can be (10s kHz+), many of those can be replaced with RCs and op-amps, too.) Then you apply feedback between the bias signal and the measured resistance signal, and have an oscillator (of sorts -- again, depending on what aspect of the frequency response you wish to select).
Probably an even more optimized (but less easily analyzed or synthesized) version can be made simply by monitoring voltage and current as you propose. The catch is, you're measuring the thermal impedance times the bias, while the bias is changing wildly. It's a nonlinear system, and likely to produce a shitty waveform in a poorly stable limit cycle -- it wouldn't be a very good frequency reference. But perhaps even more intriguing is the possibility of more complex chaotic behavior (aperiodic oscillation, even strange attractors?) by driving up the gain, and coupling the variables between more elements.
Which again, on a more general note, you'd be using the lamp as a time-dependent mixer function -- the temperature is a function of time and e.g. the square of the sum of voltages applied to it, T(t, (v1 + v2)^2), while the current depends on the sum of voltages and the temperature, I(v1 + v2, T). This isn't a very provocative function (you get the square, plus whatever T(P) heat dissipation curve you get from the mostly radiative cooling, plus the not-quite-linear tempco of the metal), but by subtracting out the linear terms you get product and square terms, and by applying (positive) feedback you can get higher order terms as well. The simplest chaotic systems only need a few state variables and a few products, so it seems feasible to me.
Another catch that may be interesting: frequency response varies with temperature, because the thermal mass remains about the same* while the heat dissipation rises quartically with temperature (radiation). This should be a very much unconventional VCO if you build it.

*It does rise with temperature, but not by very much until the melting point. We're talking an incredibly wide range here, if we're talking tungsten filaments in light bulbs!

Tim