Can you please run it again and monitor YARN UI if any job stays in ACCEPTED state for longer time (~1-2 mins)
I still feel that its RM who is taking time to schedule the jobs, because you have noticed similar behavior with Oozie and Shell script.
@Rinku Singh - I don't think this is issue at Oozie level. When you run job in parallel, it gets scheduled on available node-managers (based on data locality). In your case, we have seen that scheduling is not a problem, resources are also available, Can you please check if there is any slow node-manager? if that is the case then this issue can occur. Just try to observe common pattern by comparing multiple runs.