OK, the RUN light is now working. Those DEC engineers are perhaps a bit more clever than I gave them credit for. The AF1 signal on the bus is, well it's not bussed together across card slots. Slot 1 connects this line to the front panel connector, which drives the RUN light, which means that the processor needs to go in slot 1 if you want the run light to work.
But that wasn't the problem.
On my board, the SRUN L signal from the front panel connector drives a 7407 open collector gate, which drives the LED. I, uh... forgot/neglected/failed to connect the lead from the front panel connector to the input of the 7407 on my board. Having now wired it, the RUN light comes on at the appropriate time.
So what's the deal with it measuring 3.5VDC? Well, I suppose I should have looked at it with an oscilloscope rather than a DC voltmeter. It's not constantly on (pulled to ground), so I expect that it was acting rather like a PWM signal to the voltmeter...