would a diode between each potential(!!) voltage source and Vcc do the trick?
Not wasted I think, just cut the 5v line from the FTDI chip to the rest of the board and isolate the AVR TX signal. Also adding a connection as dc42 suggested is good if you want to detect the USB plug in.
Should I be adding diodes on all three signal lines (TX, RX, DTR) between the FTDI chip and the Atmel to prevent the Atmel back feeding power back into the FTDI?
And what about using the FTDI's 3V3OUT pin connected to one of the Atmel's analog pins as a way to detect when a USB cable's been plugged in? Since the FTDI only powers up when the USB cable gets plugged in, the 3V3OUT pin would only go high when it does. So technically I should be able to read the analog pin and figure out if it's zero (OFF) or higher (ON).
I would use resistors, not diodes, to give protection the other way round as well. 10K should be about right.
I've never added a diode on the reset line on any of my previous designs. On the MCU side, there's a 10K pullup on it. Same design as on a LilyPad which has the FTDI header broken out.