Precisely! That's why I said that the claimed behavior would require existence of an absolute reference frame. In other words, the paper Face posted is barking up the wrong tree. Conservation of energy is not the problem here, aether is.
The conservation of energy is the problem!
The energy is measured relative to the frame in which the device was initially stationary... That's the problem! The device would need to produce less and less thrust relative to that frame and it can't do that.
But you don't need to measure energy relative to the initial frame, you can measure it relative to a frame in which the device is already moving. In that case, the energy in the device changes differently with time and the acceleration decreases differently.
So we're talking about two initial frames, one moving relative to the other which would see the device as decreasing its thrust purely because of its velocity.
You would see that at velocities not approaching the speed of light yet.