Easiest example I can think of would be a method that sends some content to a JMS server. If you are in the context of a transaction, you want the message to be coupled to the transaction scope. But if there isn’t already a transaction running, why bother calling the transaction server also and starting one just to do a one-off message?
Remember these can be declared on an API as well as an implementation. So even if there isn’t much difference for your use case between putting it there and putting nothing there, it still adds value for an API author to specify what operations are able to be included in a transaction, as opposed to operations that perhaps call an external system that does not participate in transactions.
This is of course in a JTA context. The feature doesn’t really have much practical use in a system where transactions are limited to resource-local physical database transactions.