Fixed in the next release. This is in part due to the highly popular method of using preg_replace in order to handle WYSIWYG parsing as a string, rather than the less error-prone method of creating a DOM tree and adding/removing nodes. Proper DOM tree handling would be able to track current open tags and fake markup that has already been added, rather than trying to guess using string patterns. However, XenForo is designed with preg_replace and it works fine enough in regular posts that it wouldn't care to make this design change.
This creates a conflict because VaultWiki defers line-break conversion until after all other BB-Codes have been parsed (to account for templates, etc), which screws up the expected output for the step of the WYSIWYG parser that strips fake markup. However, not only is this consideration not necessary in WYSIWYG (nested templates would never be parsed in WYSIWYG mode), but this practice is error-prone already as it breaks at least the PHP tag that I have seen. Further, deferment should not be necessary since VaultWiki uses fake tags to embed parser states for the compile step, so it can do the same to suppress later line-break conversion if it was already performed.
The change is already applied on your site. Please let me know if you notice any unexpected side effects in other parsed code.