In the current implementation, we create a Properties object, then iterate over that to create a map that we store away internally (VM.savedProperties), then we remove a few internal properties from the properties object.

By first creating the HashMap we achieve two things:
 1) iterating over a HashMap.entrySet() is cheaper than iterating over CHM.entrySet() (less allocations, fewer classes loaded in a sensitive bootstrap phase)
 2) we avoid the overhead of first adding then filtering out the properties we don't want to publish

All in all a small but measurable startup gain.