I use simulators as a guideline - I never take the results as gospel. There are too many other things about the circuit, the layout, and the components that are NOT modeled and have a huge impact on whether you can get to .00x% distortion or not. A decent model will help get out major problems, assess stabillity, compare topologies, etc.
The NPN and PNP from most "complementary pairs" of power transistors aren't very well matched. Not even the modern ones.
parallel, use an 8A driver, and a triple in the first place. People always bash 3055's about poor SOA and that's not really the problem. The load on the driver typically is. An MJE1503x's are CHEAP so I don't see what the big deal is.
An EF3 would be possible using the 3055/2955 complementary. Even with a poorer match it will lilkely give a better result than the all NPN, just because it will give you less stability problems that you have to overcompensate for. The only real isssue is bias tracking but there are ways of fixing that too.
2N3773 that used a triple and 6 in parallel. Worked "ok" with a normal power supply - just a bit temperamental but eventually stable. Hooked up the class H rail switches and EVERY time the negaive rail switched in, it broke into uncontrollable destructive oscillations. Tried every imaginable compensation scheme - and no dice.
In a fit I finally replaced the negative bank with a bunch of MJ15016's, turned it on, and it's been working flawlessly ever since.
Not too long ago, I put together a little test circuit running 3 pair of 3055's in quasi triple, running off a 20-0-20 trafo. Got 120W in 2 ohms with useable SQ. Then built a complementary version right next to it. The quasi and the comp EF3 worked equally well with no oscillation problems, but the quasi needed a bit more bias to get rid of the crossover distortion.
How many trannies did you blow??
You mean you kept half of the 3773s and used them with the MJs? Are they that close to be used together?.