The workarounds are interesting: link
Both workarounds would appear to do stuff that retriggers HAL detection. So something about the config must get broken, and only HAL redetection is needed to bring the VMs back to life. Now I'm really curious what the underlying bug/problem is.