As far as I understand the system:
1) You can import graphics of any size, and resize it in image/sprite editor - the change is irreversible, and the resulting size will determine the memory requirements of the sprite, the performance and the quality.
2) You can import e.g. a larger graphics and resize the instance size in Construct window/main editor. This change is reversible - it is just scaled in viewport, and this change, as far as I understand, has much smaller influence on the performance. If you scale down the sprite this way, it still takes all the memory required by the real size of the image (=the size in the image/sprite editor), but as it is smaller on the screen, the processing of each frame is faster.