An answer has been found for the causes by @kgoebber and posted at https://github.com/Unidata/MetPy/issues/3790#issuecomment-2807622049.

The cause for both the unreasonably large negative and positive CAPE values were from the mixing-ratio calculation at levels above at least 10 hPa levels, where normal sounding would not be able to reach. By limiting the calculations to levels below 100 hPa (kgoebber) or 50 hPa (my test), the unreasonable values disappear.