In high-speed network environments, for example, networks that handle 10 GB, there’s a theoretical throughput of one billion bytes per second. A modern CPU, such as Intel® i7 or similar, can compute 160 billion instructions per second, depending on architecture and many other factors. There are 160 CPU instructions that can be executed for the entire operating system for every byte that comes in on a fast CPU. Given that a CPU must often wait, and must process other instructions, such as Interrupts and DPCs, it’s unlikely to ever reach any given theoretical maximum. This fact is why physical Network Interface Cards (NICs) use Direct Memory Access (DMA) or Hardware (HW) resources and bypass the CPU. The CPUs can't process the data that the NIC can transfer.

The problem can also be compounded by virtualizing CPUs or NICs. In modern IT environments, virtualization provides many benefits. But virtualization can contribute to the network problem because it adds another layer of processing. Thus, any perceived slowness is possibly less pronounced or absent when trying to replicate a particular scenario on actual physical hardware.

For a VMware reference that covers the impact of virtual machine memory and CPU resource limits as they pertain to scheduling of physical resources to virtual resources, see:
Impact of virtual machine memory and CPU resource limits.