Oracle VM performance issues on OVS 2.2 - java 64bit (jre 1.6) applications

Question

In an effort to move to pvt cloud & thanks to Oracle's database licensing policy on vm's we have decided to use Oracle Virtual Server as the hypervisor, OEL5+ as the database OS, and 11GR2 Grid Single Node RAC with ASM as the database. Our application can be on RHEL5+ or OEL5+ doesnt matter.
Here is my configuration (at present all componets running on only 1 physical server) :
1. OVS 2.2 on BL465G6 with 32G memory, primary storage repository is off fiber channel from a 3PAR array.
2. 40G individual LUNs physical disks presented to all application servers (Now i was using shared virtual disks & image files with file: driver and then blktap driver, for both performance sucks with file:driver being worst). I have tried OEL5+ templates, with RHEL5+ para virtualized OS as the current one.
3. On database side --> ASM disks are physical as well (just converted). Overall, the only filebased images off the repository is one of the middleware servers (MQ) , loadbalancer (Zeus) and Routing Software (Vyatta)
We are running a loadrunner test and here are the following results.
Perf App : physical linux box running appl
Perf DB : physical solaris box running db
Cloud App : RHEL5 paravirtualized vm on phy: disk
Cloud DB : Oracle 11GR2 1 node Rac , Grid Installation , ASM disks are physical , OCR/Vote disks are physical
Perf App with Perf DB     Cloud App with Perf DB     Cloud App with Cloud DB
140.08s     219.432s     226.476s
Also, i have noticed during the test only 1 vCPU is utilized. remaining stay idle.
I have run several disk benchmarking tests to determine that phy: disk presented is the best option for reads and writes. I have also run tcpdump and removed the loadbalancer and routing software out of the equation by creating a flat single subnet network amongst the appl and dB (both running on the same physical server). Tcpdump shows not much fragmentation or re-transmission and shows a similar pattern what we would see in the physical environment. The application version/ java & jre version is same on both the vm and physical server. Only difference is dB (in physical world it runs on Solaris, here it is running on OEL vm)
Note : On the OEL templates NUMA was turned off. turning that ON yielded an overall performance gain of 30%. Now i need to fine tune further. So my questions based upon my observations are :
1. vCPU calibration : Can i pin 1 physical CPU to 1 vCPU ? How does this scheduling happen or controlled.
2. Why is only vCPU showing utilization while all others stay idle. I have 4 to 6 vCPUs per vm
3. Any tuning to be done at the OVS layer?
4. Any tuning to be done at the vm's. (Note in our physical environment, apart from hardening the OS and stripping off unnecesarry rpm's we do not modify the kernel neither do we modify any TCP and/or memory buffer parameters)
Oracle support is spending days trying to figure out my issue. I have uploaded OSWatcher data so many times. I need to escalate the case but i have always received better support & information, yet quick from the forum. Any pointers would be helpful.

Answer 1

What are you missing?
I inherited this app and signing the third party jars is how it was setup, I was wondering the same thing too, why was it necessary to sign the third party jars?
The applet runs in either JRE 1.6.0_13 or JRE 1.6.0_27 depending on the other Java apps the user uses. JRE 1.6.0_13 does not have the mixed code security (so it is like is disable), but JRE 1.6.0_27 does have the mixed code security and the applet will not launch with mixed code security enable, so we have to disable it. With all the hacking going on in the last two years, is important to improve security; so this is a must.
Yes, I always clear up the cache.
Any idea on how to resolve this problem?