KVM can be paravirtualised
KVM is full virtualisation, but paravirtualised drivers for Linux's virtio subsystem (which KVM uses) are available for Windows clients (network, disk and graphics) and for Linux clients (everything).
The real advantage of KVM is that it runs atop an operating system rather than atop a hypervisor. In the original IBM VM a hypervisor made sense, as the batch-processing MVS was not the sort of operating system you'd want to use to host a VM and so some new underlying operating system was needed. There was no point adding any feature to that operating system other than the ability to run a client VM. For modern operating systems there is no good reason to replace them with a hypervisor and running the clients atop a real operating system brings all the sophistication of an operating system along for no development effort.
A secondary advantage is that KVM makes full use of the Linux's virtualisation features. These are product agnostic, and so makes for a safe long-run choice.
For all of that, KVM is not quite ready for prime time deployment. It's now a lot more stable than in the past, but there's still a lot of polish which needs to be applied. I'd say within a year KVM will be the obvious choice when considering virtualisation under Linux. This will also be good for Linux desktop users, since there's no reason beyond marketing for KVM not to be available for them. Having a sandbox available is useful for quite a few desktop applications -- such as software testing and QA and training.
Red Hat splitting its software offerings into VM and non-VM versions is more to do with revenue and marketing than with the technology.