Thanks! Looking at it from a standardization specs perspective, I would group them in the following buckets:
Features that are specified but not yet supported: OffscreenCanvas, createImageBitmap, SVG to 2D canvas, WebGL2. There is not much that can be done standardization-wise, apart from making sure that features do not end up in a spec unless they have reasonable support among implementors. This is also where it can be useful to document pains and needs in a Games CG report to set priorities.
Features that break across releases: Better test suites might help although I suspect that, in most of these cases, testing the feature in isolation (which is what web platform test suites do) is not going to reveal much.
Performance issues: Specs typically don't include performance requirements. They can still set expectations and clarify when e.g. a feature no longer makes sense if performance is poor. One recent example is the addition of "and ideally within 20ms" for the firing of cue events during media playback (see HTML spec). That may not seem like a huge change, but discussions on this triggered work to improve some implementations that were regularly firing these events every 250ms or so. Similar updates may be useful in other specs or for other features. Collecting performance issues, as done here, seems like a good starting point
Hosted runtime definition: The WebView runtime is not specified anywhere. There are more and more games that ship either wrapped in a native shim, or within hosting applications (e.g. Facebook, WeChat). In some cases, the runtime does use Web technologies but is not even based on WebView or only expose a small number of APIs (e.g. only Canvas, no DOM). In practice, that means that all runtimes are slightly different from each other. It may be worth discussing ways to converge on a common games runtime.
Support for specific formats/codecs: Specs often don't mandate support for specific audio, video, image, texture formats, because these formats may not have good royalty free guarantees. But that can perhaps be revisited. For instance, Opus is indeed mandated in WebRTC (see RFC 7874).