This is a result of one of these, or a combination
  002b7a2b * Give the worker for an INLINABLE function a suitably-phased Activation
  ca666b8b * When finding loop breakers, distinguish INLINE from INLINEABLE
  a98c9c5e * Fix a bug in CSE, for INLINE/INLNEABLE things

Some changes are quite big: for bytes_allocated we have
  T6048: 13% below expected
  T5837: 15% below expected
  T3064:  5% below expected

Of course, these might have already been close to their lower
threshold, so perhaps not all the improvement is from here.
But it is good news all the same.