I've stayed away from resonators and internal RC oscillators to avoid errors, since some of my stuff operates at temperature extremes (>100F in a vehicle in the Alabama summer, down to -70C in a high altitude balloon payload).
Most crystals/resonators are specked to nearly 160F, 100F is actually fairly close to the centre of the spec range. It's worth noting, that of the crystals and resonators I order, all of the resonators have a wider temperature range spec than all of the crystals. One resonator is specced for 145c (nearly 300F); the highest rated crystal is good to only 85c.
-70c is well beyond the operating range of most electronics, so I have no real comment on that. -40c is the lowest spec I see for both crystals and resonators. Don't most high-alt balloons include thermal insulation or even heating? Certainly all of the commercial high-alt weather gear I've seen does.
Is it greater than double the worst case resonator error?
The datasheet shows a recommended +/-2.0 error rate for regular speed clocking, the worst you can do for resonator error is 1.1%, but keep in mind, to have double this, you need to keep one resonator at -40 and the other at +85. Even then you're below the max total error rate. Even at double speed, you're below the maximum allowed error rate.
If, however, you're really operating at -70 then at over +85, you're going to have bigger issues than serial com's with the resonator, the ATmega itself will need replacing, the FT232 it's talking to will need replacing, and the regulator will too... ...all of these will need to be extended temp devices, and even in this case, the usart error is still within the maximum allowable failure (and keep in mind, this is with two resonators, that both barely pass initial quality controls, in oposite directions, and have 10+ years of service life each, decaying in opposite directions). Given that even concocting this situation, we're still within the allowed error rates, I think it's safe to say a resonator is OK for a-sync serial comms. Keep in mind, any sort of synchronous serial is unaffected by these errors, so you have to be actually using the UART serial port at these temperatures.
I would suggest you try a resonator in your target application, unless it's time keeping I suspect you'll be pleasantly surprised.