First: Thanks for all the safety warnings! I do appreciate them. I think I’m already OK with that part (it’ll be plastic wrapped and then put in a plastic enclosure where I need it).
Second: Regarding a transformer: This is a space issue, not a cost issue. A depletion mode FET costs more than a full-on regulator would! Plus it’s a learning issue – I’d like to know if this will work, or if the magic smoke will escape while on the bench (or in the wild). With surface mount components, this circuit could be a lot smaller than a transformer based one.
Third: Regarding polarity: I’m not in the general habit of tying low-power circuits to mains ground, and don’t plan on doing it here. I don’t think there’s a “right” or “wrong” polarity here, because the pins will switch galvanic connection with the phase of the AC supply power. When it’s in up-swing, the bridge connects the dead wire to ground; when it’s in down-swing, the bridge connects the hot wire to ground.
Even when using an isolation transformer, you have a similar problem, AFAICT, unless you use a center-tap transformer. (And two rectifier bridges, I guess? Which I’ve seen no power supply actually do, perhaps because I don’t have that kind of budget when shopping 
Fourth: 120 V AC is 120 V RMS positive-to-negative-swing. In the US, this is delivered as two counter-phase live wires at 240 V (or, more usually, 235V, and 117V, respectively) and dead (zero) wire is derived from ground on the site IIRC. (Europe gets three-phase at something like 400V RMS, lucky bastards The RMS for a pure sine wave means that the peak voltage is square root of two higher, IIRC – this means peak might be about 117*1.41 or about 165V. The rectifier then cuts that in half, so peak-to-peak voltage would be about 83V, and a non-sinusoidal shape after rectification.
The fact is, my circuit depends on this! The chopped power availability, forcing an “off” every 8 milliseconds (120 Hz) means that it will have to reset whatever charge is built up in the FET gates. Similarly, because there is a small amount of time where the voltage in is lower than 5.5V, the capacitor needs to be able to power the load through that interval. The values of the capacitor and the resistors are chosen with this behavior in mind.
However, the circuit does depend on actually staying in oscillation, rather than finding an equilibrium. If it reaches equilibrium, the depletion-mode FET will just act as a burn-off resistor, and quickly overheat, plus the circuit will be burning much more than the desired < 1W, leading to terrible inefficiency. In the best case, the circuit gyrates between something like 5.4V and 5.6V, behaving a little bit like a switching power supply. Except instead of a shaped square wave, it just derives the on/off cycle from load and voltage drop. I also depend on the capacitor to “short out” the higher supply voltage while being charged, so I don’t see 80V on the output terminals. The Zener will then ensure that the voltage is limited, because when it’s too high, the FETS conspire to turn off the power.
So, the real question remains: Any chance this will work? Or will it reach an intermediate equilibrium and just burn up? Or will the capacitor not be good enough at filtering/pulling down the voltage while charging, until Q1 starts telling Q2 to turn off?
