There's a lot of low-level interfaces and design decisions that filter through Thunderbird, for example networking interfaces. Once you talk about moving to a different rendering engine for the UI, you're more or less talking about ripping out ¾ of the backend code or so anyways.

Right, but isn't the argument quite a few people are making here that moving to a different UI rendering engine doesn't seem to be obviously justified, given the goals of the project and its current situation? If the architecture is as inter-dependent and non-modular as your comments suggest, wouldn't that make changing significant parts of it even less likely to be successful?

At that kind of level, you might be better putting the whole project into legacy mode with emergency updates only, and just letting it carry on for as long as it's useful and then die of natural causes when the time is right. For the other goals mentioned in the original proposal we were talking about, if there's sufficient interest to support doing that much work then an entirely new project with no baggage at all seems like a safer and more realistic bet.