Don't forget that LR is a non-destructive editor. This means that for all sources, whether RAW or JPEG, LR does not alter the original in any way. It does this partially by building a database of the image itself, so that it renders changes on screen, mimicking how they would look in the final export.

LR builds several previews, stored in the preview database. It does 'normal' previews when you import the image, but if you zoom in, the render may not have enough data to show the zoom properly, thus LR may have to do another render with more detail. These are much larger in file size and take longer to build. One option is to have LR build these on import, which will result in longer import times, but no need to wait for LR to render an higher quality preview.

Finally, I believe this is the real answer:

With RAW, there is only a small, low quality JPEG preview within the RAW file available, so LR has to create better versions for its use. As you observe, for JPEG originals, it would be possible to simply use the JPEG file itself, rather than render a preview. But that would mean that LR must have two different behaviors and methods: one for RAW, and a different one for JPEG. Very inefficient.

Using the JPEG as the source for a preview also means that you must have access to that JPEG at all times. Certain functions in LR work without access to any images. Since LR is using its database, the original can be offline, or elsewhere, and Library functions are still possible...if you have the needed previews. However, Develop module features are not available without the original.