Given the frequency with which 'home routers' have to be rebooted, I suspect ECC RAM might be quite desirable for a firewall; this may be a non-sequitur, as other causes are possible, but cumulative one bit errors may well have noticeable effects on function.
While I have a strong preference for ECC, and I agree that cumulative one bit errors could be undesirable...
I doubt any consumer-grade gear uses ECC in such a role, and I'll bet that a good chunk of "pro"-grade gear doesn't either. Most of the data on a NAT device is moving through the network and is protected by, for example, TCP checksums, and the network is typically moderately resilient to a bit corruption here or there. Bit corruption in the executable code or internal data structures could result in crashing (maybe of a service, maybe of the box) but the damage is also quite possibly going to be to something relatively irrelevant.
The frequency with which home routers have to be rebooted has a lot more to do with overall code quality than with ECC. Consumer grade devices are designed with the cheapest hardware, and have firmware written by a team of angry coders who've been given half the hardware resources that they'd prefer to have as a bare minimum, writing code under deadline, and who then are often shuffled onto another project once the crapNAT has shipped, so the remaining one poor guy who has been retained to "support" the product gets burnt out and quits, and there's no more updates, or, worse, updates from "contractors" who are brought in to try to "fix" a "critical" problem. The prime concern is getting the crapNAT out the door and onto the shelves, which makes the money, after which point, the user is stuck with the device, and the company has very little incentive to support the device further. So what if it crashes frequently?
When you look at something like pfSense (or, yay, FreeNAS!), the project works the other way around... it is the platform itself that users are buying into, and the platform continues to mature and evolve, developed by the same group of developers. Sure, iXsystems is doing it in order to sell TrueNAS boxes, but that product isn't seen as a "release-and-forget-about-it" product line.
If you look at the experience of the crews releasing software like OpenWRT, you can see that it's possible to release high quality software for networking devices. They're also well aware of the tradeoffs device manufacturers have made, and owners of these devices have figured out
how to work around at least some of those.
I hope this gives you some idea of why I don't think ECC is a significant problem for these devices. I certainly think that you can reduce the failure modes and get a more reliable device by using high quality software, and a well designed hardware platform (which can include ECC).