Hi,
What you are seeing is expected behavior. When processing content in slides, your code must take into account that text might be split into multiple runs.
The same situation exists in WordprocessingML – you must be prepared for a paragraph to contain multiple runs, and you do not know where splits will or will not come.
This, for instance, presents challenges when attempting to search and replace text. See the screen-casts at the following link, which explains how I address this issue when using a regular expression to search for text, and that text might be split across runs.
OpenXmlRegex Developer Center
Cheers, Eric