Implementing an identical canvas API on skia is an interesting idea. I wonder how the performance is compared to regular web canvas.
As for performance, Window.js has a bit less overhead since it doesn't have a DOM and doesn't implement web security.
On the other hand, Chrome is widely used and has a super talented team that knows systems programming, GPU programming, etc in a lot of depth and have spent a ton of time optimizing web APIs. So I wouldn't be surprised if Chrome and Electron have better performance in benchmarks -- this is definitely something to measure!
As an example, this p5js example runs very slowly on Window.js:
https://p5js.org/examples/simulate-smokeparticles.html
The reason is that it's doing something like a getPixel(x, y) call in a loop, which Chrome somehow optimizes; in Window.js this is going to be a roundtrip to the GPU for each pixel, which is extremely expensive.
(IMO, the right way to implement getPixel based on the Canvas API would be to read all the pixels once via canvas.getImageData)
Also invalidate the snapshot on any calls that invalidate the contents of the snapshot of course (stroke, fill, drawImage, fillRect, strokeRect... probably a couple others).
I have also reimplemented the canvas 2d api using skia for my own stuff.