Although the M2’s heated build platform works well enough, somebody who knows what he’s doing (you know who you are: thanks!) sent me an improved version. It’s a PCB heater, laid out to compensate for the usual edge cooling, firmly attached to a tempered glass plate with genuine 3M thermally conductive tape:
They designed the heater around the 30 VDC power supply used in their other equipment. Although I had high moderate hopes that a boost power supply would convert the 24 V supply I already had for the stepper driver bricks into the 30 V for the heater, it was not to be. So there’s a 36 V 9.7 A 350 W supply arcing around the planet that (I think) should work better: adjust the voltage down as far as it’ll go, soak up another few volts in the solid-state relay, and Things Should Be Close Enough to 30 V. One can buy a genuine 30 V supply, but it costs surprisingly more than either 24 V or 36 V supplies on the surplus / eBay market and won’t really provide the proper voltage without upward tweaking anyway.
I replaced their standard 0.156 inch square terminals with Anderson Powerpoles, soldered a length of shielded cable to the 100 kΩ thermistor pads, and gimmicked up a connection to the 24 V supply; it delivered 23.7 V at the PCB terminals. The thermistor is 100 kΩ at 25 °C and 11.4 kΩ at 77 °C. The PCB heater is 5.9 Ω at 25 °C and 7.3 Ω at 77 °C; it dissipates 77 W at 77 °C (no, that’s not a typo).
The ultimate temperature looks to be about 90 °C with a 24 V supply, which isn’t quite enough for ABS (which I’m not using in the M2 right now, but probably will eventually). The time constant, assuming the 1-e-1 point is 66 °C, works out to about 9 minutes; it’ll be up to final temperature in half an hour. Those numbers aren’t quite as accurate as one might wish, because the heater power drops as the temperature rises and the copper resistance increases.
A 30 V supply would dissipate 120 W at 77 °C and rumor has it that the ultimate temperature is around 125 °C, which would be fine for ABS. Goosing the power a bit would produce more heat, but I’v been running the Thing-O-Matic at 110 °C and that’s good enough. More power, of course, gets it to the temperature setpoint faster, which is probably a Very Good Thing.
Obviously, you need PWM to control the temperature; given a 9 minute time constant, a bang-bang controller will work perfectly well.
The original data, including the thermistor resistance after I got my act together, plus a cute little temperature-vs-time graph:
The colored flyspecks are part of the paper; I salvaged a stack of fancy menu cards from a trash can and padded them up as geek scratch paper.
The Smell of Molten Projects in the Morning 
I like the online graph paper generator – I didn’t know about it when I made mine: http://www.vitriol.com/images/tech/coils/Stanwyck_LHO_2.jpeg.
What minimum time-constant to period ratio is recommended for a bang-bang controller?
The general idea is that the temperature can’t change very much from sample to sample, even with the heater running at full throttle, so all the fancy control theory doesn’t buy you much. A factor of ten probably gets you pretty close; with a time constant of 10 minutes = 600 s, you can sample it every 6 s for a factor of 100.
It may take the same trickery to avoid initial overshoot as with a fancy PID controller: run full throttle until the temperature gets within a few degrees of the setpoint, let it coast while the glass plate and sensor catch up with each other, then switch to bang-bang control. Something like that, anyway.
The MOSFET temperature block used a PI control loop to compensate for the heat loss, so I may be overoptimistic.