Running Automation scripts produces vastly different results on Linux and Windows

[martinwinter-huawei]

The following videos have been captured on Windows 11 and Arch Linux, both running the same code base (both o3de and o3de-atom-sampleviewer on the latest development) and produce vastly different results.
The video shows running the shadowtest automation script suite on both platforms:

Linux:

VideoLinux.mp4

Windows:

VideoWindows.mp4

Not only is the runtime completely different, but also the viewport scales differently and all test cases fail (compared to just two on Windows).
Note that directly starting the Samples/Features/Shadow testcase works and looks the same on both platforms, these discrepancies only show when using the automated test scripts.