It will never be as fast as Sublime. Not before a few years at least.
It uses regex with some off-the-wall engine that they chose because it supports just about every encoding under the sun. Many of the language syntax packages I looked at could easily improve their regex performance.
Plus when you open a file as "plain text" or disable the syntax highlighting package much of the sluggishness goes away.
I'd love to dive in and try and fix it myself but I won't have time to for a few months. The code itself isn't horrible just a little undocumented.
My personal guess is that is because Code might use a more stream-like processing of the source files and does not have the whole files in memory (it also works with really big files - which Atom does not support) and that this does work better
You are probably right about the streaming system. It's interesting that atom doesn't do it that way (seeing as how streams are such an useful part of "server-side" javascript)
Like I said at one point i'd really like to dive into this stuff and spend some time trying to figure out exactly what the problem is (and maybe help find a way to fix it).