This ought to be impossible. The pin in question hasn't changed between the two revs of the Uno, nor does it have any relevant alternate functions. (Analog comparitor, IFF configured, one of the pins involved in parallel HV programming, which requires ~12V on RESET.)
The only thing I can think of is that the D7 trace goes "close" to the resonator pads (but in the other side of the board!)
If the resonator was somehow "iffy" (the normal fuse setting is for a "low power crystal", and I had at least one resonator on an ATmega8 board that insisted on "full swing oscillator), having a solid signal on that wire might change the behavior. But ... then I'm not sure why it would work after a manual reset (well, Vcc would presumably be more stable...)
Another possibility is that the AVR is somehow mis-progammed, or a new chip revision with bugs, or something like that.
If you take the AVR out of one of the new and misbehaving boards, and put it in a Rev1 board, does it start behaving?
Do you have a way of reading the fuses?