Keeping high benchmarks dispute started by VMware in its comparison between ESX Server and Xen, which XenSource tried to close providing a much fairer comparison between ESX server and XenEnterprise, OpenVZ published on its blog a performance analysis of OpenVZ patches for Linux kernel 2.6.16 against Xen 3.0.2, both on Fedora Core 5.
The original work, a 130-pages diploma thesis by Björn Gross-Hohnacker, is in German language, but OpenVZ team has been so kind to translate in English 17 pages of it, from which we can read following conclusion:
The comparison of both processes led to the result that OpenVZ can deliver an extremely efficient virtualization approach for a deployment within a Linux operating system layer. This approach offers a level of performance similar to a standalone physical system without a perceptible loss through the implementation itself. On the other hand, Xen offers the functionality of a very comprehensive resource allocation offering the opportunity to exclusively make a large part of the virtual components available to guests, thus ensuring high scalability. In summary, one can state that the analysis and evaluation of the different implementations showed that no virtualization process is optimal for each kind of use case. It is important to first conduct an extensive survey for each application in order to select a solution which is most suitable for the respective requirements.
On these statements OpenVZ team provided an additional reading:
IPC and disk I/O performance is better (or much better) for OpenVZ than Xen, CPU-intensive tasks are about the same for both, networking is a bit better in OpenVZ. Conclusion: for homogeneous (i. e. Linux-only) environments, OpenVZ is way better — as it was designed to be.