The first mention of Nested Page Tables (NPT) can be tracked back to the end of 2006, when both Intel and AMD disclosed their plans to introduce the technology in future versions of their CPUs.
Processors makers and virtualization vendors promised a remarkable performance boost thanks to this technology, at the point that VMware even stated that the virtualization overhead could be completely eliminated by 2010.
Until few months ago there was no chance to verify the claims.
The first CPU to offer a NPT implementation was AMD in September 2007 with its new Quad-Core Opteron (formerly Barcellona) and the Rapid Virtualization Indexing (RVI) extension.
Intel will not follow until Q3 2009, with its upcoming codename Nehalem CPU and the Extended Page Tables (EPT) technology.
The hypervisors that support AMD RVI at the moment are VMware ESX 3.5, any commercial product based on Xen 3.2.0, and KVM.
AnandTech just published a short but very interesting preview of their analysis, showing the performance improvements obtained by Novell SUSE Enterprise Linux 10 (which implements Xen 3.2.0) running on a 2-way system with AMD Quad-Core Opteron CPUs at 2.3 GHz.
They configured 4 virtual machines with 2 vCPUs and 4GB RAM each to run Windows Server 2003 R2 and the following workloads: 2 IIS web servers serving PHP, 1 Oracle OLTP and 1 MySQL databases.
The benchmark highlighted a performance boost going from 7% (Oracle) to 31% (IIS with PHP).