If you just use the standard Mendix fuctionality, you are safe from XPath injection as far as I know. User input will be escaped when inserted into an XPath query that is generated by the framework, and all instance access will be applied to the query as well. Even if you manage to send a custom XPath retrieve request to the Mendix server, entity access will still be applied and it is not possible to retrieve any data you are not authorized to see (if you have configured entity access properly). The same reasoning goes for updates and deletes.
Of course, if you are generating XPath queries in custom Java in an insecure way (i.e. string concatenation without escaping), this does not apply. In that case, you are responsible for the security yourself.
If (despite the above) you have a demonstrable testcase of an attack where data can be illegally retrieved or modified, you should file a support ticket or contact your partner manager ASAP.
Also a good best practice is to keep all entities manipulated (or even read-only) by the "Anonymous" users in-memory (non-persistable), else it could be possible to (XPATH) query those entities via XHR call manipulation.
Although the Mendix Business Server to some extent provides a number of layers of security (such as SQL Injection) X-Path injection should be avoided by setting X-Path constraints.
It is fine if users are able to alter the xpath queries, that shouldn't cause any harm as long as your domain model security is setup properly to make sure no non accessible results are returned.