When I started writing this post, the title was "Benchmarking Hyper-V". But I've quickly realized the limitations of the few tests I've collected. Anyway, here they are.
The first limitation of my test is clear: I should have tested Hyper-V against VMWare Server, not against the Player.
The second limitation is the test itself. I quickly looked for something that supported multi-cores, ran standalone and presented fast results. All I could find was Orthos. Orthis is yet another prime solver tester (a version of Prime 95), and it is good enough for my first "kind-of" benchmark.
So here are the results (1 minute CPU test):
VMWare: 281.75 GFLOP
Hyper-V: 514 GFLOP (245 + 269.5)
Ok, not fair, I should really test against VMWare Server, where I could run 2 cores... But this is what I really use. I'm virtualizing my desktop, and VMWare Server video driver is too sluggish to use as desktop - though Hyper-V is not significantly better...
I expected more from Hyper-V, being para-virtualized. Though not as fast per core, Hyper-V is clearly faster using my DualCore. And it runs smoothly on my Windows Server 2008 - VMWare Player needs 64bit driver signature recognition to be shut of.
So, and for the time being, I'm porting my VMs to HyperV. Until I really need a lot of memory, then I will probably run back to VMWare, who knows.
A last curiosity: for those like me that are used to application virtualization, having both cores at 100% on the guest and both idle on the host is strange. Understandable, yet strange.