I'm pretty convinced this is because the app switcher can recognize _whether_ a part of an app's UI is a native text field and not just an image, but it's not smart enough to pull the text out. I've seen it make number 0 -> letter O and other such mistakes before, so I'm convinced it's using OCR. Which is probably harder for Chinese.

It's crazy to me that it can't just use the accessibility APIs.