There are SQL statements available in the oracle performance tuning guide to get the average process time of each BPEL process – 


select composite_name Composite, component_name Process, componenttype, state,
       count(*) Count,
      trunc(Max(extract(day    from (modify_date-creation_date))*24*60*60 + 
                extract(hour   from (modify_date-creation_date))*60*60 + 
                extract(minute from (modify_date-creation_date))*60 + 
                extract(second from (modify_date-creation_date))),4) MaxTime, 
      trunc(Min(extract(day    from (modify_date-creation_date))*24*60*60 + 
                extract(hour   from (modify_date-creation_date))*60*60 +  
                extract(minute from (modify_date-creation_date))*60 + 
                extract(second from (modify_date-creation_date))),4) MinTime, 
      trunc(AVG(extract(day    from (modify_date-creation_date))*24*60*60 +  
                extract(hour   from (modify_date-creation_date))*60*60 +  
                extract(minute from (modify_date-creation_date))*60 + 
                extract(second from (modify_date-creation_date))),4) AvgTime       
       from NTSOADEV_SOAINFRA.cube_instance
       where creation_date >= to_timestamp('2014-03-05 20:08:00','YYYY-MM-DD HH24:MI:SS')
          and creation_date <= to_timestamp('2014-03-05 20:11:55','YYYY-MM-DD HH24:MI:SS')
        group by composite_name, component_name, componenttype, state
          order by count(*) desc;