> >  4. Look at your power supply filter cap size and its effect on
> >     Vdd ramp up time.
> The power is coming from a handheld computer, and is very clean and well
> defined, so I don't think that's the trouble.

Taking regged power across a cable is generally a mistake.
Measure it at the chip, you may be suprised.

> Again, 3 boards work fine in exactly the same conditions, so it must be
> something on this particular board.

Or just a marginal design. Happens all the time.
I've seen 10 protos working fine, and when the design was put into
production, 50% failed. (A PHD engineer put external loads on a modem chip's
osc that wasn't specified to drive any external load.. "But all the protos
are working!"..... )

>This one was reworked a little more
> than the others, so maybe an iffy solder joint or a stray bit of grunge?
> Where should I look, on the OSC pins?

You might try lifting the xtal/resonator up a bit. Depending on the package,
you can get a little crud caught under the package, which could be causing
the problem.