A.Thread management inside the JVM process might be less effective than context switching between processes on the OS level. It is better to combine both these technologies to achieve full saturation of resources.
B.Current CPUs are multicore units. So starting managed servers in the amount of physical cores and pinning them 1:1 to cores will always get the best performance.
C.By a combination of prioritization of processes on the OS level and Work Manager in WLS, one could tune the whole environment more precisely.
D.Assigning more than one managed server to hardware will never bring better performance.
E.More instances means more connections to databases, so one could serve more concurrent calls on the database level. This is primarily why this approach will give better performance numbers.
F.Depending on the profile of the application, having more JVMs with smaller heaps will lead to more predictable response times than fewer JVMs with a large heap.
