Okay, we have finally made progress; and what results we have, my friends! Thinking that there definitely seemed to be something hardware related to my problem, since the problem would migrate as I connect other pins and subsequently set them as outputs, I played with it a bit more. I tried manually pulling the modem ~reset low by shorting the appropriate test pin (Pin 71 on the MC1321x) to ground. Strangely enough, the debugger died, saying that the MCU reset was being pulled low. By finding that out, and a bit more playing and using my powers of deduction, I was able to get the processor to initialize correctly, and even get an interrupt from the modem, presumably meaning the modem is working correctly (I have yet to do any work with it yet after figuring this out). To get it to work, I set the corresponding bit in the PortD data register so that the line would be held in the not reset position. Then, I could set the pin as output without a problem.
In other words, the modem reset is tied to the entire processor's reset, and by pulling the pin low (or initializing the port as output without first setting the data register) I was effectively executing a hard reset in software. This pin was acting like a big red self-destruct button that I ritualistically pushed during processor initialization.
Now, the questions. Why? To again quote the documentation that I quoted in my original post,
"From a reset condition, the MCU initiates PTD3 as a high-impedance input with the internal pull-up disabled. As a result, the modem reset input will be floating and the modem will not be held to reset. As part of the MCU initialization PTD3 must be programmed as an output and then driven low to reset the modem. The ~M_RST input is asynchronous and needs to be held low for only a short period." -MC1321xRM, section 3.4.1.2
It doesn't make sense to me that the documentation would be so incredibly wrong, or that the modem reset would be tied to the processor reset. I thought it might be attributed to bad soldering, but the pins in question are on opposite sides of the chip - there would be plenty of other places that would short before those particular pins would short. _However,_ further examination of the underside of a board located a via just barely shorted to a trace, connecting the ~RESET and ~M_RST networks. Problem solved.