Thank you for the cable-routing tip...
I was trying to decide whether to mount the header on the CPU side of the board or the backside of the board:
The CPU side had the advantage of not having to worry if the header left enough room to shut the case and easy cable routing out the back (there's a screw hole right above the 'func' button), but the disadvantage of harder access and possibly contact with hot silicon on the business side of the board.
The backside had the advantage of easy access to remove the serial cable and avoidance of hot ICs on the front of the board, but the disadvantage of not having a nice place to route the cable out of (but routing it out the front sounds good), and the possible issue of clearance of the header vs. the case.
If the header fits on the back of the board and lets me close the case I'll probably do the same as you. BTW, have you had any noise on the serial lines? At least the NSLU2 forums suggest possible AC hum on the lines keeping the machine from botting and I'm wondering if it's worth worrying about now.
As for reflashing, I found what appears to be the problem... I was giving it the initrd.img (and renaming it as initrd.buffalo) vs. extracting the initrd.buffalo from the poorly-cloaked zip-as-img file. Oops, somehow that was never clear in the fragmented docs I gathered up.
I'll try to reflash it before soldering anything on the board, but since I have the headers and have a PL2303 cable already sacrificed for the job, that will probably happen soon anyway.