I can reproduce the bug on multiple (physical) servers running in our data center.
When booting my virtual servers with the new kernel, the issue did not arise... ...until I enabled NUMA. Booting a virtual machine with 2 sockets and 16 cores each works fine when NUMA is disabled in KVM. With NUMA enabled the machine boots only if I set the number of sockets to "1". When I set the number of sockets to two, the machine crashes.