Ive tried to get rid of min_size by using just one of the 2 allocation strategies unconditional, but for both variants there is significant speedloss This patchset is for documentation and should not be applied (unless theres some flaw in the testing which indicates that its better ...)