I've also been encountering this problem -- so I decided to investigate it in a bit more depth and record a video as well as data output, in the hope that this will shed some light.
The following video shows some trial runs:
This is with X-Plane 11.21r2, Saab 340A v1.5, and all plugins except Gizmo64 were disabled by moving them to a different folder.
The video begins right after X-Plane has finished loading, to show the startup in case I'm doing something wrong. (I'm using the Auto-Start feature.)
In the trial runs, to keep it simple, I'm merely doing some high-speed taxis on the runway: Bring the engines to a high power setting (CTOT not used), accelerate to 80 knots, then bring the throttles to reverse and decelerate.
The video shows three runs, starting at the following times:
Run 1: 4:32
Run 2: 5:35
Run 3: 7:10
On the first two runs, I bring back the throttles very slowly and deliberately. I encounter no issues in these two runs. However, the throttle movement also seems excessively slow to me -- slower than I would expect on a rejected takeoff, for example. On Run 1, it takes 7 seconds from high power to reverse; on Run 2, 5 seconds.
On Run 3, I bring the throttles back more quickly. However, I wouldn't say I'm yanking them; it still takes 2 seconds from high power to reverse. On this run, the engines spool down shortly after the throttles reach the reverse range -- note turbine temperatures and RPMs.
There aren't any temperature or RPM exceedances.
At 7:35, shortly after the throttles reach the reverse range, the mixture on both engines goes to 0.0 -- see "mixt1" and "mixt2" data readout at the top left. I don't have any joystick axes configured for mixture controls, so something in the aircraft logic must be commanding the mixture to zero.
The mixture going to zero appears to be what causes the engines to spool down.
Even more interestingly, at 7:37, the mixture goes back to 1.0, but at this time, the engines have already flamed out.
Two seconds from high power to reverse doesn't seem excessively fast to me -- am I wrong? At least I'm surprised that it would cause a flameout. The fact that the mixture briefly goes to zero looks a bit suspicious to me -- maybe there's something in the airplane logic that could be fixed to prevent this?
I'm attaching my Log.txt, GizmoLog.txt, and a Data.txt that I captured during this run.