Enabling Stylus access to all sites tanks performance in Speedometer and other Browser performance websites

[nintendork07]

Description

1. Allow Stylus access to all websites in a chromium based browser like Brave ("Allow this extension to read and change all your data on websites you visit: All Sites")
2. Go to Speedometer 2.0 or 3.0
3. Test with the extension enabled and disabled. On my Apple M1, the score should be 32 for Speedometer 3.0 and 435 for Speedometer 2.0. When the extension is enabled, I only get 22 in Speedometer 3.0 and 285 for Speedometer 2.0.

It can be fixed if you disable Stylus OR if you allow Stylus access to ONLY the website you use it on (for me, I use it only for youtube so I put it to ONLY allow Stylus access to youtube's website). Speedometer gets normal scores then. This issue has been occurring for me for the past few months, but I just now figured out that the Stylus extension was causing it.

System Information

• OS: MacOS 15.3
• Browser: Brave 1.75.175 - Chromium 133
• Stylus Version: 2.3.10

Screenshots, links, CSS

Expected score

Score with Stylus enabled with access to all sites

I only have one style enabled in Stylus for youtube called AdashimaaTube.

https://github.com/sapondanaisriwan/AdashimaaTube

[tophf]

I only see 1% or less difference in Chromium browsers on Windows.

[nintendork07]

I'm seeing quite a huge difference on Windows 11 as well. On my 9800X3D, I'm getting a baseline speedometer 2.0 score of 541. With stylus enabled, I get 487, with subsequent scores of 384 and finally 322. This is especially noticeable if you run them back to back.

With the extension disabled, it stays within margin of error around 540, but with stylus enabled it's a large dropoff.

It only happens with this one extension, even if I disable all others.

It occurs on both windows and mac, and I've tested both microsoft edge & brave. All have the same issue.

On windows, it seems speedometer 3.0 isn't affected but on mac it definitely is. Speedometer 2.0 is affected on both. If I set stylus to only have access to one site (youtube in my case) then the scores remain unaffected.

[tophf]

Well, I can't replicate it, so maybe you could record a trace yourself in devtools -> Performance -> ⟳ button, then upload it here.

[tophf]

Could you test the nightly build? I've disabled one potentially expensive call.