overdue timers are fired automatically after session has been restored.
To know what sessions need to be restored would depend on your session management strategy. What you could do is to try to group sessions and those with timers would belong to given group that might be identified by name - the grouping mapping would need to be outside of the runtime engine itself. That way when the process goes into timer node you'll have way to find out what session might be worth restoring. Alternatively to timer events you could use signal events and the ejb timer will do the work (signal ksession or process instance) based on time intervals - since you already plan to use ejb timers. So the ejb timer would simple look through the grouping/mapping to find what is awaiting signal and just load the given session and then send signal on it.
HTH