On some platforms, transparent huge page support can be advantageous to KVM performance: http://www.linux-kvm.org/wiki/images/9/9e/2010-forum-thp.pdf Currently, the only way to have qemu benefit is to unconditionally enable THP support on the host with the 'always' policy but there are valid reasons not to want to do that. A better approach is to have the application itself use madvise(2) to to enable huge page support for any allocated memory regions as appropriate. This support has finally made its way into qemu as of commit ad0b532 (a one-liner patch). If there are any plans to stabilize >=qemu-1.2 in the near future, I would really appreciate it if this patch were to be considered for inclusion.
This is in qemu-1.2.2