All the time, every time for the past 3 years. Have completed clusters with 10+ systems in this way staggered over a couple of hours unattended.
As an estimate we would have done this probably 200+ times, and have only ever had one system fail miserably that had one of the defective "Renice" SSD disks installed so would have died at a general reboot anyway.
Others have failed for different resaons (alarms, post upgrade checks etc) but always roll back to the previous release and identify in the log what the failure reason was to look at afterward.
We have a public facing FTP we use for this purpose for all our maintained customers.
The painful part is you need to have the software in a directory structure based on the release you are upgrading from.
Eg. Currently on 7.1.2.24, it will look in MCD-3300/7/1/2/24 for the FTP software.
Our server is a linux platform, so I wrote a script that creates a structure of sym-links that all reference back to the same SW-ROOT folder.
We can then upload any current version to the SW-ROOT folder and know that any system we point to the FTP will get that version.
It works awesome.