because it is well documented that Smart Objects can be endlessly rescaled non-destructively.
No, you can not. Scaling any pixel data follows the same rules, whether that is an SO or a conventional layer. Scaling "endlessly" is only possible with vector objects in both directions, but even their rasterization follows specific rules so quality may degrade under some circumstances. For pixel content, only scaling down is "lossless" with SOs, but the data will still be resampled which may be perceived as a loss in quality. And well, scaling up is gonna look just as unglorious as with any otehr layer. Of course, you can always reset to original size, but there's no magic beyond that.
There is much modern focus on "non-destructive" editing, but keep in mind if you don't overwrite or destroy the original file there is no destruction at the highest level. Put in layman's terms, you could always start over with the raw file.
That thought segways into my next one: Non-destructive editing makes sense if you need to use the same information for a variety of somewhat related purposes, or if the work product may need to change (e.g., to suit the whims of a fickle client).
But at another extreme, if you're editing for a particular purpose - say creating the best possible print from an exposure - sprinting right for the finish line by changing pixel values directly and being done with it can be an extremely effective approach. This requires that you get things right the first time, and that takes practice.
Some folks do their Photoshop work by building up layer after layer and using smart objects, smart filters, etc., and this can be effective but no computer has yet been built that can composite all that stuff in real time with a big image. So there IS a cost to doing it. What you might gain by being able to re-do things, you might not have needed to gain if your control responses were instantaneous and you could tweak the intermediate result at every step very easily. Note the number of posts about how slow Photoshop CS6 is/was at editing deep documents, some by people using 2012 computers.
As with most things, it's horses for courses. It's good that Photoshop gives us rich tools and choices for how to work.
Regarding your specific question, bear in mind that what's communicated to the parent document from each of its embedded Smart Objects is a flat, rasterized image. Think of the embedded smart object kind of like going off and opening another document, making the changes you want, saving the document, then flattening it and pasting the pixels into your parent document.
In the very first example in the linked article, they show how the smart-object-rasterized image of a vector circle, subsequently scaled by resampling the parent document in which the Smart Object is used, becomes fuzzy as it is scaled up. Once you understand this you realize that of course you could scale up the smart object itself, e.g., to a size equal to or larger than what's ultimately needed by the parent document, and then it could be crisp in the parent document where it's used.
Of course, having all your smart objects at a size larger than you need takes up even more resources.
Oh, and by the way, I tend to do the "sprint to the finish line" thing more often with my own photo editing. I actually LIKE starting over from the raw file, and I find advantages in doing so:
1. Technology changes. Photoshop CS6 and Camera Raw 7 make better results from my raw files from the past decade than any prior software has been able to do. So starting over gives me a better quality result. New filters in Photoshop proper are available (e.g., HDR Toning) that weren't available years ago.
2. I change. I learn new techniques, and my tastes change. So starting over allows me to choose different steps. You could say this could be a good example for doing it the "smart object, layered" way, but I might counter with this: In order to rework an existing document, I would have to "figure out" how I had put the old document together, then reorganize/change it. That might limit the results. I find it advantageous to meet each new image editing challenge afresh. Then I'm not shackled to something I did before. My style can change more freely.
3. I actually very much like editing images in Photoshop, and making an even better result this time than I did last time is gratifying.
That's not to say I don't use smart objects and layers where they're warranted, for the reasons I gave near the prior post. But I NEVER let the word "destructive" flavor my thinking (save for the fact that I make darned sure I don't lose my original files). For me it's more about the word "creative".
Resizing scaling smart object layers can also be done by editing the embedded smart object and using Image size to take advantage of vector and text layer that may be in smart objects because these type of layer are resize using vector graphics and changing font sizes not by interpolating pixels. Additionally layer styles can also be scaled. Here is a link to an old thread in the read only Photoshop Windows forum. http://forums.adobe.com/message/3499623#3499623