I would contend that 3*drivers image (post #89) is a similar setup as neilki's (post#91) insofar as they are upsets arising from an unusual attitude commencement. 3*'s in particular is not likely the consequence of a stall event in the sim, resulting in the attitude. It is correct that some time ago the 737 sim did have tweaking to the aerosim model, and that has resulted in a more realistic stall and post stall behaviour. Buffet in the stall of the B737 specifically is most pronounced at the low altitude stall cases, either clean or configured for landing; the high altitude stalls and accelerated stalls are quite gentle in comparison. High speed buffet is a fairly high frequency vibration that increases in intensity dependent on increasing speed and g load/aoa. High altitude stalls give a mild low frequency buffet that is not as unpleasant as the stall buffet at low to mid altitude and configured for landing. The 737 meets the rolloff requirements of FAR in almost all cases, unless the slats have been poorly rigged, then a rolloff at the break or just before the break can be quite interesting.
Years ago, while evaluating a 3rd party TCI and his students, the NG crew kept on planting the box in the dirt following an encounter into an aggressive microburst model. While they were recuperating and licking their wounds, I ran a quick QTG eval, and then ran the same scenario. The off axis entry into the microburst was causing a pitch up, which is correct, and a yaw towards the center of the microburst, which is also correct. The net result was the crew were adding a pitch up with the initial warning, which was taking the box into a stall while there was a lot of yaw rate, and the box was doing a 180 roll and being planted. If the crew respected the PLI (aoa limit) or kept the ball in the center, the box behaved impeccably. The TCI wanted to fault the box for it's behaviour, until he was debriefed as to the QTG eval and the importance of aoa and yaw management in a stall. Again, the sim even in the out of balance stall was providing reasonable modelling, and was still recoverable, but it needed to be flown to recover from what should be an expected outcome of a stall with a yaw rate on. This sim model was in stark contrast to another 3 engine model that could be fully stalled and would climb in stall at 6000FPM, whereas in the certification testing the plane had demonstrated normal stall break.
The difference between the actual aircraft to the simulator is the continued period of a load experienced at the seat, and the intensity of the buffet that can be encountered. In most cases the real aircraft will leave you in no doubt as to the fact you have stalled the aircraft at the lower levels.
The observations above come from a rather misspent career. Flight crew use SOPs and the FCOM/POH to avoid getting into these areas, which is highly appreciated by the SLF and their families. The planes however behave like a variant of a Cessna 150 out on the edges of the envelope. They certainly behave better than the PT-22 in a stall following Hap Arnolds design input into what had been a beautiful design.