One will end up creating a segment and another will trigger putting the segment to disk. The limit essentially denotes the amount of memory that we can afford for a particular shard in memory- which makes sense.
Is there any other reason to keep them the same? I am sorry but I could not find any documentation on this.