My guess:

Without zoom, a raw picture is taken at resolution x by y. The picture is compressed to jpg introducing some jpeg artifacts.

With zoom, a raw picture is taken at resolution x by y. The picture is cropped to (x - z), (y - z). That image is then scaled up to x by y, losing raw image quality since some pixels are doubled and others might be averages etc. The picture is compressed to jpg, introducing a different set of jpeg artifacts.

I don't think that there would actually be more detail there, but in some cases the application of the jpg algoriithm would be less effective at compressing the image, in which case it would have to leave more of the raw data and not introduce artifacts. The file size will likely be larger (at least for that section of the image, other areas may balance out).

You can probably test further by setting the jpeg compression level to 100 (which I believe is no compression, though it might still apply some even with this setting), and trying your experiment again.
You can set the jpeg compression level in an ini file somewhere - there are a few references somewhere on this forum, but from what I've read the current settings produce a good quality to filesize ratio.