One thing I found makes a huge difference on performance is Power management settings for the server hardware and/or OS, that slow down the CPU speeds to save power when CPU utilization is low (which are often enabled by default). On a modern system, VDBench doesn't consume enough CPU to cause them to run at full speed, so the may be running at say 35%-50% of full clock speed. I see huge differences in throughputs if I test with the power setting set to static high performance mode vs not, especialy with the smaller IOs (higher IOPS workloads).
I did check here internally, and that is indeed something that has been reported by others.
- make sure that you run long enough to allow the cpus to 'warmup', give the OS time enough to realize that it needs to ramp up the cpus.
- deactivate power saving mode during performance testing.