"How to Manage ZFS Dat...

You are joking? Proof of concept? You know that a legion of people and companies use this in production? In Solaris 11 you can't even boot from something different?

Both points you mention are inherent to the inner mechanisms of copy-on-write filesystems as you surely know. When you don't overwrite data in an existing object in your dataset, you have to write it elsewhere nescessarily. Thus on rotating rust devices this may leads to the IOPS inflation problem when you update your files frequently in the sense that you just change blocks in a file. On SSD this is a non-factor as i recently proofed again in a customer project.

You further surely know, that you can switch the "percentage of free space"-value when Solaris switches from first fit to best fit for new blocks which is a feasible strategy when you know there are just a few files acting as a container for the i/o load of an application.

And as a side node: You never ever fill up any filesystem with a lot of of write load over 80% or 90% (except you put just create once some files in that filesystem acting as containers for the i/o of an application) as almost all filesystem run into some strange performance pathologies over this level under really high load.