This package contains a small server and Angular application to non-rigorously measure speed differences between REST-like calls to a server made over (a) plain old AJAX HTTP requests versus (b) a WebSocket connection.
To fire up the server.
npm install imperfect-rest-comparison node imperfect-rest-comparison
Then launch a browser and open up:
You'll find a page with options to run various tests against the server.
If you've managed to tweak your browser to turn off the Connection: Keep-Alive header in AJAX requests, then HTTP is going to be very slow in comparison to Socket.IO. This is pretty easy to do in Firefox if you are set up to use it as a developer, less easy in Chrome.
The unscientific results for browsers with Keep-Alive on, using non-cached AJAX requests and non-cached equivalent WebSocket exchanges:
The story no doubt becomes different if you're making heavy use of client-side caching. Good caches for AJAX HTTP come built in to frameworks like AngularJS and are essentially free in terms of development time. You'd have to implement something similar from scratch for your potential WebSocket transport layer. Not that this would be an enormous task. Still, something to think about.