High memory use of embedded web tabs

I run Remote Desktop Manager 2026.3.15 / 2026.3.16 on a few Windows 11 ARM64 VMs (Parallels on Apple Silicon, 8-10 GB RAM). The VMs are used only for RDM, with several Website entries opened as embedded tabs. Most of the memory and CPU goes to the embedded browser, and I would like to understand what RDM does there and what I can safely tune.

What I measured:

- Right after start RDM has about 10 msedgewebview2 processes below it (1 browser, 4 renderers, 1 GPU, 3 utility, 1 crashpad). After two days with the tabs open it is 13 processes using about 2.1 GB, and RDM itself is at about 0.9 GB working set (574 MB private) with roughly 4,500 handles. On 2026.3.15 I saw RDM growing by about 5.6 GB over roughly 4 days.
- One renderer hosting a KVM video (WebRTC) tab used about 12.6 hours of CPU time in 2 days while the stream was open. A tab with a busy web app (an RMM web console) kept another renderer at about 40 % of one core plus about 20 % in the GPU process, around the clock, while nobody touched it.
- The WebView2.Cache folder was 13.4 GB on disk: about 100 stale WV2Profile_* folders (about 11 GB) in an old EBWebView folder that is no longer used, plus a separate user data folder for each distinct browser-argument set.
- Updating the Evergreen WebView2 Runtime from 145 to 156 reduced RDM's WebView2 private memory by about 120 MB (604 to 482 MB). Adding --disable-features=SpareRendererForSitePerProcess through the AdditionalBrowserArguments policy for RemoteDesktopManager.exe gave nothing measurable (same renderer count, memory within noise).

Questions:

1. Does RDM set CoreWebView2.MemoryUsageTargetLevel to Low, or call TrySuspendAsync, on web tabs that are not visible? If not, is that something you could add, perhaps as an option per entry or globally?

2. Is one CoreWebView2Environment shared by all web entries, or does RDM create one per distinct browser-argument set or profile? In my case each argument set seems to get its own browser process tree. Can all entries be made to share one environment?

3. Does RDM ever clean up or cap the per-entry WV2Profile_* folders under WebView2.Cache? Is it safe to rename or delete the old folder while RDM is closed, and what is lost besides saved web logins and cookies?

4. Is there a supported way to limit the number of renderer processes or to suspend idle tabs, and are there browser arguments you support for web entries? Does RDM honour the WEBVIEW2_ADDITIONAL_BROWSER_ARGUMENTS environment variable?

5. Which entry types or panels are known to keep a renderer busy in the background (web-app logins, WebRTC/KVM video, the dashboard)? I have not verified this, but would the "Use embedded Microsoft Edge for overview panel" setting under File > Settings > Advanced make a difference to memory or CPU?

6. Are there known memory leaks in RDM itself that are still open after 2026.3.16, or fixed in a later build? If it helps, what should I capture (dump at a given size, ETW trace, logs)?

7. For a VM dedicated to RDM with no real GPU, which settings do you recommend to keep memory and CPU low (rendering and performance options, which WebView2 runtime version, how many web tabs)?

Thanks a lot for any hints. Happy to share more numbers or test a build if that helps.