My guess is that it's almost impossible for the programmers (however talented) to 'reproduce' stalls in a simulator;
Not being argumentative, but this is not quite correct.
Flight test data from stall testing, to the extent it was conducted, is incorporated in advanced Full Flight Simulators. These flight tests are obviously structured for some measure of safety, so that beyond a certain point where data collection is out-weighed by risk, any additional data used in simulation is predicted data from other sources.
Beyond the stall boundary there is still a certain amount of training value where simulator characteristics and response to pilot inputs is valid. The problem is you cannot necessarily tell where flight test data transitions to predicted or extrapolated data during a more extended encounter. And truthfully this has not been given any priority in regulatory evaluation requirements of individual simulators. You don't "down" a sim for something for which you have no experience or guidance for subjective evaluation or a flight validated QTG for an automated evaluation.
The big jets are stalled and recovered so as to produce stall warning system functionality and tolerances and recommended avoidance and recovery procedures and technique. Simulations don't discount this.
As has been stated previously, however, improper pilot inputs to any simulation can also take you out of the realm of what was tested and into "new territory".
The advantage of simulation, of course, is that pilots can make improper inputs and get negative feedback safely. There is currently a lot of effort being directed at substantially improving jet transport stall simulations and incorporation into operator training.