A few months late but for those who might run into similar problems I wanted to follow-up on this with what we found and how. Believe me, this is the highly condensed version.
As mentioned already, multiple people tried their hand at troubleshooting this thing including a "30 year Mazda master tech". Each repeated and reaffirmed the prior work and occasionally tried something new but to no avail. One thing that kept coming up was the immobilizer which I felt we had eliminated, but the way that the ECU responded to the procedure for reprogramming it was particularly odd and inexplicable so we couldn't rule it out conclusively.
I finally installed a Megasquirt PNP because it eliminates the immobilizer crap and has extensive logging capabilities. With much cranking and sputtering the car would occasionally start but run like crap with lots of sync errors and VVT errors. Monitoring the VVT position vs VVT maps it was clear that the VVT would get stuck in a position and the only way to break it loose again was to ramp the actuator duty cycle way up and then abruptly drop it back down. Then it would work until the next time the engine was stopped. While it was working it ran fine even on the dyno. This seemed to make sense because it explained both the sync VVT errors. So, repair or replace the VVT mechanism and related pieces and we should be good. Well, no. I did find that the VVT actuator had been apart before and reassembled incorrectly so I cleaned it up and put it back together. The car still had bad sync errors. Opened it up again and compared the intake cam and cam position trigger lobes to a spare we had. There was no obvious damage or conspicuous difference in orientation and the lobe ring was tight on the cam. But the spare VVT mechanism was significantly different internally, and I think a better design (I'm guessing from a later year). So I decided to swap that in, but because of the differences I also needed to swap the cam gear which of course meant pulling the belt again. At that point I figured we may as well pull the intake cam for a much closer look. Bingo! When I placed the cam next to the spare in the identical position, the sensor pickup lobes were in a slightly different position. And when I looked real closely I could see that on the one we took out it had not been pressed fully onto the back of the cam. I'm told that the stock ECU won't tolerate moving that very much and still pass the sync test, which makes sense. Add to that a VVT mechanism that may get stuck in the wrong position and maybe an old stretched belt, and things are too far out so the ECU just doesn't allow spark. Unfortunate the ECU also did not provide any useful clues in the form of fault codes. The Megasquirt doesn't care about any of that if you don't want it to so we could manipulate things on the fly while logging all the sensors and that helped us to focus attention in the right area. Once we had it back together correctly it fired right up and ran perfectly though numerous dyno tests on both the Megasquirt and the stock ECU.
This car was built from new as an SSB for a prior owner of this site and has passed through several owners and shops since then, so there is no way of knowing when or by whom, but somebody decided to cheat a little on cam timing by moving the trigger lobes. That might seem a little odd on a car that already has VVT but it's still designed to work within a range intended for the street with compromises not ideal for the track. I seriously doubt it made a big difference but as usual with these cars it is many small things that add rather than one big conspicuous cheat.
So, recommendation to SM tech for next year, develop a simple way to check the trigger lobe orientation on the intakes cams. Let me know if you need more information or help.