Whenever we decide to use Compressed Oops with 64Gb heap AT LEAST next changes should be made to build VM which support it.
Note: more complex changes needed to use a VM option. I don't see benefits on N2 with it so I am putting it on shelf. 

src/share/vm/utilities/globalDefinitions.hpp
@@ -288,11 +288,11
 // Minimum is max(BytesPerLong, BytesPerDouble, BytesPerOop) / HeapWordSize, so jlong, jdouble and
 // reference fields can be naturally aligned.
 
-const int MinObjAlignment            = HeapWordsPerLong;
+const int MinObjAlignment            = HeapWordsPerLong * 2; // *2 for 16 bytes alignment
 const int MinObjAlignmentInBytes     = MinObjAlignment * HeapWordSize;
 const int MinObjAlignmentInBytesMask = MinObjAlignmentInBytes - 1;
 
-const int LogMinObjAlignment         = LogHeapWordsPerLong;
+const int LogMinObjAlignment         = LogHeapWordsPerLong + 1;  // +1 for 16 bytes alignment
 const int LogMinObjAlignmentInBytes  = LogMinObjAlignment + LogHeapWordSize;