The first recording sound like theres something really wrong in the accumulator->waveform generator of voice2 and lesser so in voice3. Some waveforms work and others do not. Its not the accumulator itself since the triangle2 doesn't work while the saw2 does. According to Yannes, the triangle is made from sending the accum to EXOR gates, this probably whats at fault in your chip. Never heard that yet in any of my tests
Exactly what failed would need a die scan

But safe to say that this SID is toast.
Though that 'gurgling' on the _PS_ waveform, I've heard many times in lots of chips and in my examples.