Note that when I posted my response above, I was running 1862.2. Continuing with that, this is what I've found so far:
* With 1862.2:
** Waited for an hour or more (lunchtime) after the kernel debugger should have connected. During that time, an initial connection was made, but not completed. The debug target is still hanging before the Windows boot screen. In other words, I believe that the serial port stuff "works", but it's full of so many errors that it is extremely slow and unusable.
** Changed baud rate to 19200 from 115200. This didn't seem to help.
* Then I downgraded to build 1848.
** Tried to connect with kernel debugger at 19200. Still no success, after waiting for 15 minutes. The debug target is hung before Windows boot, and the debugger says "Waiting to reconnect..." There is some traffic on the serial port, as it is flashing...
** Used Hyperterminal in both VM's. Now at 115200, I can communicate across both without errors.
** Tried to connect with kernel debugger at 115200. Still no success, after waiting for 10 minutes. Again, the debug target is hung before Windows boot, and the debugger says "Waiting to reconnect..." There is some traffic on the serial port, as it is flashing...
I seem to have found that the Beta 1862.2 makes the virtual serial port worse, with the communication errors that I saw. I could not replicate those comm errors in 1848. However, it still doesn't work. I think that I'm doing something wrong with Windows Kernel debugging, but when I did this on VMware with my Linux machine with two WIndows virtual guests, it always just worked.