For some reason that is currently unknown to me (did not investigated) this instances are much better with Redis since they are able to fork using more or less

For some reason that is currently unknown to me (did not investigated) this instances are much better with Redis since they are able to fork using more or less 1/20th of the time needed before. (4.4 GB process forked into 44ms instead of 1 second). EDIT: it's just hardware virtualized mode (HVM).