Never mind, I think I've fingered it out.
The reason of the quality drops was exporting the 720p video to 1080p. When stretching an image bigger it is unavoidable that the result become a bit blurry. Even when you stretching the 1080p image back to 720p you can't get the image as sharp as it's original. This problem is more noticeable when you have text or other sharp elements in the video.
If you want to keep the the text as sharp as the original video, it would be better to avoid stretching by exporting to the same resolution. Please see below for my testing.
1. Comparing exported 1080p (left) with the orignal (right). You can see obvious difference:

2. Comparing exported 720p (left) with the original (right):

For H264 compressor, you don't need to set the quality to highest. The second image above is exported using default quality.