Yeah, you need good soldering to get good soldering, of course.
Also an option, use a riser board to handle the control circuits, etc. This can be whatever, and a big bulky header carries signals into the heavy board -- this solves problems like SMT trace/space limitations, and saves precious board area.
Also good practice to simply not need traces at all, as much as possible. An inverter bridge for example, can be laid out with the transistors and connectors stacked all right next to each other, so that the current paths are as wide as the components, and less than a mm long, say. The current density might be off the charts, but it's so short that it's really pretty irrelevant. You're relying on lateral heat spreading here, not surface dissipation.
Ampacity can also be improved with thermal transfer (clamp the board between thermal pads and heatsinks?), minding that, while that saves trace temperature, you still have all those losses, which need to be dissipated for one, and also amount to whatever mV of wasted voltage.
The same is true of the previous paragraph, i.e. all those transistors, and what little connecting copper lies between them, all needs to be heatsinked.
This is getting into the regime where rules of thumb (like trace ampacity) are irrelevant, and even hand-waving calculations are questionable (estimating the voltage drop of some trace by an equivalent rectangular bounding box). You really might want to just build a 3D model and simulate it in a thermal physics sim, or budget the time and money for several protos to measure it directly, and iterate upon.
Tim