Pre-emption is by default turned on Databricks cluster. Turning on or turning off pre-emption would make more sense on a high concurrency cluster. Pre-emption ensures that the job starting for resources gets a fair share of the resource available on the cluster. To make the resources available, pre-emption would kill the running tasks.
To answer the question, as there are no concurrent jobs on an on-demand cluster, it would not make much difference if the pre-emption is enabled or disabled. In case if there are concurrent jobs submitted using internal application logic, then there will be an impact.
