How Hadoop scalabality linear or proportional depends on the number of nodes

Hi,

According to Hadoop processing capacity scales linearly, but not proportional to the number of nodes. Is there any analysis about this?