In the wake of JMM9 overhaul work [1], we need to conduct a few performance studies.
One of these studies include figuring out if enforcing the safe construction of *all* objects,
regardless of whether they contain the final fields affects performance in any sensible way.

This issue tracks the experimental VM work for that; it can be also merged to master,
if we want to simplify performance study freeing everyone from building the patched VM.

[1] http://openjdk.java.net/jeps/188