With FS-Mark we are also able to run the same 1000 file, 1MB file size test but with sync/fsync disabled, and when doing so, the performance between the four configurations was roughly the same. This confirms though that in the other disk benchmarks, VirtualBox behavior is not following sync/fsync requests by the application.
With enough disk benchmarks under the belt, the next area of focus is upon the CPU performance between the virtualization platforms. With the 7-Zip compression benchmark, KVM was running at 85% the speed of the host operating system while VirtualBox 3.2 was at about 75% the host performance and with VirtualBox 4.0 Beta 2 it regressed to being just 50% the on-metal performance.
With Parallel BZIP2 compression there was not much of a performance hit taken by the hardware-accelerated VMs, but KVM commanded a slight lead over VirtualBox.
The Linux Kernel-based Virtual Machine did wonderful with the C-Ray ray-tracing benchmark where its performance nearly matched that of the host with this multi-threaded benchmark while VirtualBox was behind by several percent.