VMXNET3 has a lot of functionality built into it that it's prior iteration (VMXNET2) doesn't seem to have however most of it seems geared to extremely high throughput environments.
It supports RSS (receiver side scaling) though according to the document I'm referencing, not in the linux kernel (though the linux kernel itself supports RSS). MSI(-x) which exponentially increases the number of interrupts available to the adapter. LRO (Large receive offload which is a much needed capability on high bandwidth production VMs' in my experience) and the New API framework for packet processing (
NAPI).
At least, that's what it says on the box.