Camera.Render has poor performance when observed in 2022.3 and later streams

How to reproduce:
1. Open the attached “camerarender_benchmark” project
2. Open the Build Settings (File > Build Settings) and ensure that “Development Build” is enabled
3. Press “Build and Run” and wait for the application to start. It will run a benchmark and automatically close after a few seconds
4. In the Editor, open the Profiler window (Window > Analysis > Profiler)
5. Press the button depicting a folder in the top right corner (looks slightly different in later versions)
6. Navigate to “Benchmarks” in the build folder (e.g. “camerarender_benchmark\Builds\Benchmarks”) and select the benchmark data to open it in the Profiler
7. Open the Profile Analyzer window (Window > Analysis > Profile Analyzer) and press the “Pull Data” button
8. In the “Name Filter” under “Filters”, enter “Camera.Render”
9. Take note of the values found under “Marker Details for currently selected range”

Expected result: When comparing results with later versions, the median value of Camera.Render (ms) per frame should stay roughly the same
Actual result: The median value of Camera.Render (ms) per frame is higher in recent versions

Reproducible in: 2022.3.22f1, 2023.2.0a17, 2023.2.1f1, 2023.2.15f1, 2023.3.0a14, 6000.0.0b12
Not reproducible in: 2021.3.36f1

Reproducible on: Windows 11
Not reproducible on: No other environments tested

Notes:

• Explanations for regressed versions and test results are found in the internal notes
• The issue seems to be GPU-dependent, with a higher (relative) discrepancy in performance between different versions on higher-end GPUs. About 50% increase from 2021.3.35f1 to 2023.3.0b6 on NVIDIA GeForce RTX 4090, and a 23% increase from 2021.3.36f1 to 2023.3.0b5 on NVIDIA GeForce GTX 1650 Ti
• Camera.Render is on both the Main thread and the Render thread. By default, the Profile Analyzer will only compare performance on the Main thread, but this can be changed under “Filter”
• This project uses BiRP and was not tested in URP or HDRP due to the way the benchmarking process is set up