Crosswalk has only been around a couple of months, and in that time it's already improved a lot and fixed a bunch of bugs. In the near future I am absolutely confident it will be bulletproof and we're working closely with Intel to support the last features everyone wants (ads, IAP etc). The fact there are bugs in the beta is to be expected and they will be resolved as Intel are actively working on it. Several issues have already been fixed which you can test already with the beta/canary build option. Given how new Crosswalk is and how steady its progress has been I do not understand why anyone would be ready to write it off already. Of course all new engines go through a beta period with a few issues before reaching maturity, that is normal. It will get there soon and the existing progress and even the current state of the canary build option is good evidence of that.
I also want to point out that we are designing the engine to scale well to large, ambitious scale games. I believe games like Airscape and Penelope prove this is already possible. Optimisations like the recent collision cells go a long way toward making this a reality (collision cells themselves being a relatively new feature). Perhaps a reason there aren't so many "big" games in Construct 2 is people think since there aren't many big games, the tool must only for small games, and then not so many developers attempt big games. I don't think this makes much sense when big games are already being developed and published with Construct 2, but the stereotype persists and probably to some extent is a self-fulfilling prophecy. We also know lots of people develop small and medium sized projects commercially for various clients including for mobile browsers, where it serves as a profitable and efficient tool. Perhaps that's just an easier way to make money than with a single huge make-or-break game; either way, we don't mind, we just make the tool. But we are still thinking about how the engine could better handle large games in case that's what people are after, and any optimisations aimed at large games will probably help smaller games too.
We also have more optimisations in the works for future builds and we know every browser maker is as conscious of performance as ever and are working on optimisations on their side as well. For example we're aware that one of Google's priorities with Chrome this year is to maximise GPU performance on mobile, and they have some very interesting optimisations in the works towards that (e.g. the ubercompositor, zero-copy rasteriser, etc). All of this will also reach Crosswalk.
As for iOS, CocoonJS does work for a lot of people for publishing iOS games. We are still trying to work with them to get support for things like memory management. If there are other issues, as ever sending them a clear and detailed bug report with a minimal project to reproduce the problem can mean the difference between getting it fixed and not getting it fixed. However as a lot of you have also mentioned, we're not entirely satisfied with Ludei's rate of progress, so to provide an alternative we will look in to an official Ejecta export option in the near future (we are currently tied up with multiplayer, but hopefully some time in the next stable cycle). However AFAIK it does not have memory management, no Web Audio API support, no WebRTC, no Web Workers (for faster pathfinding), no XML parsing, and no vibrate support. Some of these may be fixable but it remains to be seen if this will be any better than CocoonJS. Still, having a second option could be useful for some games.
I'd also point out that nobody in this thread so far appears to have tried to argue that native exporters would still be worth it given the likely disadvantages from my previous post. Having a portable engine that can guarantee that even just the logic will work 100% identically is a pretty big advantage and removes a whole layer of porting problems. Remember we are trying to pursue solutions that are both fast and compatible to make porting worry-free. Crosswalk is a perfect example of a perfect solution to that end. We are trying to resist spending months of painstaking work to trade performance problems for some very awkward compatibility problems.