Performance doesn't matter if nobody can use it, and right now nobody can use it. By the time you have a working C++ implementation, for all you know v8 and SpiderMonkey's optimizers will be generating code just as fast as your C++. You're allowing an engineer's natural love of premature optimization to distract you from actually getting your product in front of people.

C++ is also less cross-platform in this sense because to support multiple platforms, you will have to support compilation across target platforms and handle the variety of plugin APIs available for those platforms. And on iOS, your stuff won't work at all because it's Javascript or the highway if you want to run in Safari. (I guess you could ship your own browser on the App Store, but I suspect you'd run into rule hassles.) Maybe you could use Native Client, but in that case you will need to ship native client binaries for each CPU architecture and you'll only work inside Chrome.

Attempting to entirely supplant the browser's rendering engine with yours in a plugin is also a hopelessly shortsighted move. You'll be throwing out all the hardware acceleration and advanced layout logic provided by modern browsers. Just rendering text is complex enough to require multiple full-time salaried engineers, and you will have to bear that burden if you will ever be compatible with HTML/CSS, because international text is important.

Build a prototype using JavaScript, HTML, and Canvas/SVG where needed. Put it in front of people. Let web designers and programmers play with it. If they can play with it, they'll find reasons to love it. Once you have web designers and programmers who love your actual product, you have a path to success.