The fragmentation is calculated by giving value to each free chunk.
Simplified, byte chunk size X*2 is 1.1 more worth then byte in chunk size X.
This logic is not very useful when chunks are really large.
So, for large chunks (>8MB) we change value coefficient to 1.02.
Overall, the change is intended so split into large chunks will only slightly
increase fragmentation, but preserving the sharp increase of fragmentation when
small chunks start dominating.