8028993: Full collections with ParallelScavenge slower in JDK 8 compared to 7u40

Reducing the number of calls to follow_class_loader to speed up the marking phase. Also removed some unnecessary calls to adjust_klass.

Reviewed-by: stefank, jmasa, mgerdin

hotspot/src/share/vm/oops/instanceKlass.cpp

@@ -2199,7 +2199,6 @@ int InstanceKlass::oop_update_pointers(ParCompactionManager* cm, oop obj) {
     obj, \
     PSParallelCompact::adjust_pointer(p), \
     assert_is_in)
-  obj->update_header(cm);
   return size;
 }