Before jumping into the gory details I felt it would be good to give a bit more background on why the design is what it is.  Aside from the obvious specs of what it's supposed to do, the layout & format was a big deal.  The ultimate goal was to keep the physical design as simple as possible.

The TS50 offered perspective on a modular approach, which with some refinement could lead to a compact & flexible form factor.  But the TS350 has to deal with much larger currents & dissipation.  Connecting wire introduces more loss and a lot of EMI.  Board-board connectors would be better but they're expensive, complicates the physical design, etc.  So I settled on a single board.

To surface mount or not:  that was my debate for the MOSFETs.  I settled on SMD because it led to easier assembly:  no heat sinks to mount and no faraday screens. 

I had calculated a peak dissipation of 30W so the SMD approach made for a relatively simple thermal management scheme:  two fans, each mounted at the center of the board would draw air over the HV supply & sine sections and blow air on the hotter inverter section.  All MOSFETs would use SMD heat sinks.  This strategy also eliminated concerns for where & how to mount fan(s) in the enclosure.

Transformer design also drove the layout of the inverter.  Toroid and RM cores are what I had experience with and available.  The math showed that, per unit area, I could get more power from a toroid at a lower operating temperature.  The problem with toroids is finding good mounts for them, especially with many windings and high currents.  In a home brew you can cheat by mounting directly to the PCB.

As you can see from the picture or schematic, the sine output filter inductor is huge - 5.8cm in diameter.  This was necessary to meet the required 12mH inductance with a multifilar winding capable of handling RMS currents in excess of 4A.  This complicated the layout of the sine bridge while trying to keep the overall board dimensions reasonable.

It all looked good at the time.  In hindsight these compromises actually led to a clumsy layout with large current loops - more so on the sine section (hideous EMI) than inverter.  When you look at commercial & reference designs they almost always use TO220 vertical mounts.

With the redesign I'm almost certain the inverter will move to these and likely the sine too.  That sine inductor will get a size reduction thanks to a higher operating frequency and I'll probably put the relay & panel stuff on a second board.