Starting with our basic Apache server benchmark, there is a big hit taken when switching to virtualization from running this web-server on the host itself. This though is not too much of a surprise, but between KVM and VirtualBox 3.2/4.0 there is not a huge delta. VirtualBox 4.0 Beta 2 was the fastest with being approximately 12% faster than the Kernel-based Virtual Machine.
With SQLite, VirtualBox was much faster than KVM and even the host operating system, but this is particularly troubling. Running SQLite was over 80% faster running under VirtualBox than on the host operating system itself. We have seen this before... QEMU/KVM even was like this before. We ended up reporting it and getting it resolved, as Matthew Tippett talked about during the Southern California Linux Expo with The Five Stages of Benchmark Loss. At least as far as the default configuration is concerned, VirtualBox is not properly obeying disk sync requests. Therefore, VirtualBox may be reporting much faster numbers, but your data is potentially at risk in the VM with a default VirtualBox configuration.
With PostMark, VirtualBox 4.0 Beta 2 is also faster than the system host itself. It is actually "faster" than VirtualBox 3.2.12 where the appropriate behavior is to perform slower than the host is, so the default disk behavior with VirtualBox 4.0 may have actually become worse. The KVM number is appropriate at around 47% the speed of the host.
When running FS-Mark with 1000 files of 1MB size where sync/fsync is supposed to be used, VirtualBox is again much faster than the host.