Field layout code makes no use of the super-class gaps. Which means if we have the space in superclass area, we don't use it. For example:

    public static class A {
        long f;
    }

    public static class B extends A {
        long f;
    }

    public static class C extends B {
        long f;
        int a;
    }

...is laid out like this:

  jol.Main_05_InheritanceBarrier.C object internals:
   OFFSET SIZE TYPE DESCRIPTION VALUE
        0 12 (object header)
       12 4 (alignment/padding gap)
       16 8 long A.f N/A
       24 8 long B.f N/A
       32 8 long C.f N/A
       40 4 int C.a N/A
       48 4 (loss due to the next object alignment)
  Space losses: 4 bytes internal + 4 bytes external = 8 bytes total

It seems legal to lay out the fields like this (i.e. move the C.a up to fill the gap):

  jol.Main_05_InheritanceBarrier.C object internals:
   OFFSET SIZE TYPE DESCRIPTION VALUE
        0 12 (object header)
       12 4 int C.a N/A
       16 8 long A.f N/A
       24 8 long B.f N/A
       32 8 long C.f N/A
  Space losses: 0 bytes internal + 0 bytes external = 0 bytes total

That saves 8 bytes per instance. While this is arguably hard to do with current field layout code. By Java rules, C.a is not visible in A anyway, and there is no legal way to hijack its value. Am I overlooking something?