The reason why we can nondestructively scale a smart object up and down is because unlike a normal layer, Smart Object Layer saves the pixel information of an image.
Usually when scaling an image up or down, PhotoShop changes all the pixel data and then retrieving the original starting image will be virtually & physically impossible.
Not just scaling images, even when putting effects, PhotoShop will change the pixel data of the images and retrieving the original form will be very hard, if not impossible.
Imagine you are going to use this image of a racoon in your project.
If you accidentally or on testing purposes over sharpen the image and make it like this -
And save the file.
You won’t be able to get back the original image.
But if you’ve made the image a smart object layer first, you will be able to edit the sharpening or completely remove it and retrieve the original image as is.
In short, the smart object layer holds and saves all the pixel data of an image and any edit or effects you do gets added as a layer under the smart object layer. That’s why you can reshape, resize and put effects on a Smart Object without affecting or destroying the original image.