Hi
Did you remember to include gate source protection Zeners? This is very important with any mosfet that does not have them included intrinsically within the device. An unprotected gate-source junction can destroy the device without it even being in the circuit, this is what makes them so static sensitive.
First power the circuit up without the output mosfets and tie the NF point to one side of the Vbe multiplier, don't worry too much about a little DC offset at this point. Then measure the voltages where the gates are to connect. You should be able to adjust the Vbe multiplier to a range with a minimum voltage below 2 times Vgs threshold, and be able to adjust it to ~1V above 2 X Vgs threshold. Fix the circuit so you get these results
before adding the outputs. After including the missing Zeners and the missing output coil || 10R resistor, set the bias pot so the voltage between the gates will be at the minimum, i.e. below 2 times Vgs threshold, add the output mosfets and power it up with no load. Then adjust the Vbe multiplier pot until you measure ~100-200mA through R14 & R15, a respectable bias for these devices. Hopefully this will help keep you from needlessly destroying more mosfets.
As a more general side note, Trench and U-type fets do not have a very good SOA within the 'linear region'. This is primarily why they are not chosen for linear applications. It's not secondary breakdown per say, but is similar in destruction. They do have properties that make them better switchers though; class D, SMPS.
Also there are Planer Stripe Diffusion Mosfets, another type out there. Fairchild's Q-fet line is akin to these. They have a die structure forming a unified singular well as opposed to the 'Hexfet' types that have a discrete plural well die structure that forms multiple tiny transistors all connected in parallel. (see photo) From my experience the planer stripe mosfets are quite solid through the linear region and can take more abuse than the cellular types but they seem to exhibit more non-linear characteristics, particularly Gm that is a bit dependent on Vds. I believe the planer stripe mosfets can be better in terms of SOA but only if a form of local error correction is used to tame the Vgs vs Id distortion that is related to Vds, in addition to all the other distortions inherent to mosfets. (IMHO, no mosfet should be used as linear outputs without EC due mostly to their non-linear capacitances, not to say it will not work without EC, but I digress

)
I can crank out continuous 120Wrms @8R speaker load with very low distortion with only 2 pair of TO-220 Q-fet planer stripe mosfets.

I have yet to try the TO-3P, I have not yet the need for that much power in my living room.
All of the V-fet mosfets are designed by the manufacturer to be used as switchers and the datasheets are geared towards it but that doesn't mean they will not work as linear amplifiers. Some are very good if used properly.
