Comment # 12 on bug 994969 from Ruediger Oertel

well ... in short: too much extra effort to support separate virtualisation
per architecture.
our qemu/kvm works on all platforms.
For power this means on bare metal, because stacking kvm on top of powerVM
looses quite some performance of the machine and with recent kernels does
not even work (since then you need kvm-pr instead of kvm-hv, and kvm-pr has
been broken for quite a while now, actually I can only remember a handful
of kernels where it worked at all).

in a differnt approach, using plain powerVM you do not get too much further.
in the standard setup, powerVM is not scriptable, the approach with
novalink integration never finished and the bootup times of a powerVM lpar
compared to a qemu/kvm VM you loose lots of cycles again ...

hammering buildjobs on a machine is a specific workload, and qemu/kvm
on baremetal has up to now been by far the best approach.