Sree,
I've never subscribed to the 1 JVM per core philosophy... My approach is for careful evaluation of the application and it's behavior. I've found that I make better use of the hardware available to me than the 1 JVM per core approach.
During evaluation we look at the CPU Queue length, Disk I/O, network usage, and memory utilization. This gives us an idea of how to make the most of the hardware with the applications we need to host on WebSphere.
Many times, from my experience, I find that software vendors tend to state their application requirements in terms of dedicated hardware. This creates a scenario of underutilized resources and increased cost.
By profiling and distributing your applications based on compatable profiles, you can much better utilize your hardware and do not need to subscribe to the 1 JVM per core methodology. Of course, an application may require that if you're application is profiled and found to be CPU intensive.
Also consider the fact that IBM offers Cloud Burst and Hypervisor editions of WebSphere Application Server. These editions are designed to oversubscribe your hardware and go away from 1 JVM per core (in my opinion.)
Erik