There are several aspects to the question. The question talks specifically about the case of creating the inline image by uploading a new copy, from the perspective of resource usage. But asking the question also implies a question about the inherent value of inline images, and the question asks about that, too; whether it's a sufficient improvement to justify accepting the edit.
Original link vs. new image
Other answers cover the fact that there are no substantive implications for system resources. I'll suggest that not only is it not a cost or harm, it's actually a worthwhile use of system resources.
If someone is familiar with markup language, they can modify the existing link, as suggested in the question. But it's easy to make mistakes or typos when you do that, and the process can end up eating a lot of time fixing it. The editor has a provision for quickly creating the inline image.
A small minority of users volunteer their time to curate site content. It's more important to make their time more productive and less frustrating than it is to save a few bytes on a server. The time they save not having to fiddle with markup could be used for other curation, or to save time and frustration so that curation is less of a chore. If we can reduce their frustration and wasted time even a little at the virtually non-existent cost of a little server space, that is a great trade-off.
Is it an improvement and does it justify an edit?
Assuming the image adds value, I generally see enough value in inlining an image and getting rid of a link that I approve those kinds of edits as a substantive improvement of the post.
A comment on Glorfindel's answer asks "how exactly" does that make a post much better? I'll take a stab at that. As a baseline, I'm assuming it's agreed that images can illustrate or clarify a point in a post and make a big difference in people understanding what is meant by the words. You're asking only about their being in the form of a link vs. inlined. A few thoughts:
Reading continuity. People can assimilate and process information, including different formats of information, quickly if they can take it in all together as a unit. It can all go into short-term memory together, but that needs to happen within the same short time and without distraction. Once it's in short-term memory, the brain processes it, making associations and links between the elements of what was taken in and things in long-term memory.
Anything that isn't mentally-usable information (like the text of a URL), can't be incorporated. Instead, it's just noise that slows down the processing and then gets ignored, or it acts as a distraction because making a mental note of needing to go back and look at that interrupts processing of the rest. If you mentally deal with the URL, you're likely to have to reread the rest to refresh your memory. So if it's in the form of a link, it slows down and disrupts comprehension of the information. That's a long-winded way to say that it makes the post harder to read.
An image is likely to distract you, too, because you will stop reading and focus on it. But the image is a very condensed form of information; it is the information ("a picture's worth a thousand words"). So focusing on that helps make the associations to comprehend the information.
Self-contained posts. We ask users not to post a link as an answer for a number of reasons. A big part of that is that links go stale (which isn't the issue here). But it's also because we want the posts self-contained. People can read the post and get the information, they don't have to read the post to get a link that they have to click on to get the information. There's some flavor of that with image links. You don't get the information when you read the post, you have to read the post to get the link, and click on the link to get the information. It's part of making it easier to get the information, which is one of the things that makes the platform valuable.
