The site listed doesn't give a lot of information on what exactly the corpus is, but as the paper is from 2010, we can assume that there isn't quite as much bias towards social media posts as there would be if such a study occurred today. That's important because social media tends to be "written" in a register that's much more in line with the spoken register of a language compared to traditional written media.
As such, this phenomenon is not unique to Chinese, but can be noticed throughout pretty much any language with a large corpus of written media. Much like a news article in English would likely use the word "purchase" more often, whereas spoken English would tend towards "buy/bought," it would be natural for a similar situation to occur in Chinese. This would result in a different frequency distribution for characters in written Chinese as opposed to spoken Chinese.