After a sudo poweroff, the Pi3 continues to draw around 95mA.
Bar using an external HAT board eg MoPower, WittyPi et al, I took a look to see if there were any hacks to get the current draw lower.
Checking out the Pi3 partial schematic, first thought is turning the perfect diode's Q3 off so as to completely chop the power; alas the mosfet is wired "backwards" to the usual p-channel high-side switch, with the body diode pointing forwards, so no dice there.
Elsewhere, there are circuit elements to ensure that the supply rails come up in sequence highest->lowest, so as to avoid catastrophic latch-up in the SoC.
3V3 is enabled first, as the enable pin of the 2306 dual Vreg is tied directly to 5V. The enable pin for the 1V8 supply is pulled to 5V with an RC delay of ~10ms - that's next up. Lastly, there's the 1V2 core regulator supply, which is switched by Q7, whose gate only goes low ~2ms set by the RC delay on Q8's base current, which is derived from that 1V8 rail.
So immediately there are two levers to pull & investigate reductions in current draw. Care must be taken to take the rails down in the reverse order they came up, never allowing a lower rail to be present in the absense of a higher one, or bang goes your Pi.
Turning off Vcore first gets the draw down to ~50mA. That's the GPU & ARMs off.
Additionally after turning off 1V8, it sips 11mA, with ~2mA of that the red power led.
Additionally, cutting PCB traces topside so that I can pull EN1 low, so switching off 3V3, I see ~2-3mA. Strangely, both power & activity LEDs are dimly lit at this point, where I thought they'd both be off. If they were, I guess draw would be under 1mA inclusive of my RTC. 3V3 has drooped to 2V5, not fully off. Ideas?
I wouldn't recommend these techniques, frought as they are, but it's interesting to see the breakdown of standby current by power rail, and to gain an understanding of the Pi's power rail circuitry.
Edit: properly turning off Vcore, 1V8, then 3V3 results in current consumtipon <1mA.