Unfortunately, I seem to have spoken too soon. The problem is back, and it doesn't matter what power supply I use. As long as I use any power supply, be it switching or unregulated, serial communication fails after a period of between 20 minutes and 4 hours.
The failure happens either with the arduino's response not being received and its tx/rx lights staying solid for that duration, or, the com port in windows simply refuses to work anymore "A device attached to the system is not functioning." The only fix is unplugging and replugging the USB cable, and all is well again. Note, the uno does not need to be power cycled. I even put in an uptime seconds counter in my code to verify it wasn't being reset, and it isn't. So, it's a really strange problem. It only happens when using any power supply in the barrel jack. If I leave it on the usb connection for hours, there's no serial communication failure. I verified my circuit isn't causing the problem by running the trials without any connection to pins and a power supply directly into the barrel jack. At this point I am wondering if my board is somehow damaged. The 16u2 chip has rev 0001 on it, is that rev known for failures like this in windows?
It really seems like a waste to have to get an ethernet shield or some other way to communicate with the uno when there's an existing way already built in. It should work without failing like this, and requiring physical disconnection and reconnection of the USB cable to (only temporarily until it happens again) resolve.
Addendum: My code is flashed onto the 328p through the ICSP so there shouldn't be any reset problem being caused by DTR. As mentioned, I've verified via an uptime counter that the board is not being reset.