It is the worst case scenario under heavy load.
And exactly what e-mail and database servers have to cope with under really heavy load.
The ZFS filesystem groups and serializes the writes in order to write to the disks at friendliest way to rotational disks technology.
But what happens when a heavy load multithreaded random small files synchronous writes is sustained for a long time period, exceeding the ZIL VDEV size and L1 ARC size?
In order to answer this, you need to understand some of the factors that impact ZFS performance.
You need to read the bibliography at the article bottom.
http://www.techforce.com.br/news/linux_blog/zfs_part_4_sustained_random_small_files_sync_write_iops