I don't see a benchmark of network IO, especially small packet performance, or heavy random IO workloads on SSD?
If those are the kinds of apps you want to run that is great, but I would need to see more to get an idea of how they compare for frontend, caching, content, or database serving.
The difference I see between virtualization configurations for things like small packet performance is dramatic even within the same hypervisor.
That's actually really interesting. I wanted at one point to run Joyent's SmartOS (http://smartos.org/) on a desktop, since it seems like I live my work-life in VMware anyways. I'd like to know what VMware's processor cost is, because I've definitely noticed it indirectly when trying to do the same thing both inside and outside the VM. Let's take my last example: a Python script which searches a large graph to find out how many paths there are with N edges or fewer. It turned out to save a couple hours of running time to install Python on Windows and then run on Windows than to run on Linux in the VM, and I may be wrong, but I don't think that the Windows version of Python is magically faster than the Linux one.
I also haven't figured out how to get VMware to let me do OpenCL stuff, which has occasionally been a headache when some library does exactly what I want to do, but does it with OpenCL. The hope would be that bare-metal hypervisors would be a little more transparent.
But yeah, 2% is great news. I don't even consider this a "win" against Xen necessarily; I think they're both showing that they're much more competitive than I thought they'd be.
If those are the kinds of apps you want to run that is great, but I would need to see more to get an idea of how they compare for frontend, caching, content, or database serving.
The difference I see between virtualization configurations for things like small packet performance is dramatic even within the same hypervisor.