I know that running COM1 and 3, with exactly the same code does have not issues.
I do not have a logic analyzer sorry.
This was also with a series of relatively simple test code harnesses written by two people independently (as you do when you want to remove all other possible reasons for the bug).
I will see if I can get access to a logic analyzer, but in the interim everything is just chatting away over COM1 (COM3 has its pin positions covered by an XBee shied) and that seems to work for now. And yes, we removed the XBee shield in our tests!
Pete