Which framework/server/platform would you choose to build a REST API to deliver maximum performance?

For performance I would suggest nodejs, it's really really fast. Throw varnish on top of that and your golden. It also has the benefit of being able to scale much higher than ruby or python without major rewrites.  Source: Myself, but that won't do and I'm lazy so here's the first google link: http://www.infoq.com/news/2012/10/Ruby-on-Rails-Node-js-LinkedIn

Node.js with mongoDB, try this sample app http://coenraets.org/blog/2012/10/creating-a-rest-api-using-node-js-express-and-mongodb/

If raw speed is the only concern I would say is a pretty good choice for using as a backend langauage. Here are a couple of reasons. Stability its process model allows you decouple different parts of the application into processes. If part of the system fails, it doesnt effect the hole but itself. Concurrency is first class citizen in Erlang which built into language (rather than using underlying system's) and its very cheap to create couple of thousand threads. Checkout whatsapp use case = http://blog.whatsapp.com/index.php/2012/01/1-million-is-so-2011/ OTP framework, Error Handling are other nice to haves. Check this comprasion of Erlang web frameworks : https://github.com/evanmiller/ChicagoBoss/wiki/Comparison-of-Erlang-Web-Frameworks , http://www.ostinelli.net/a-comparison-between-misultin-mochiweb-cowboy-nodejs-and-tornadoweb/

There are other things to consider apart from performance. I don't think there are great differences there regarding most used technology stacks. Our blueprint is Java on Spring (planning to move to java + CXF). It lets us deploy either on IaaS or PaaS. PaaS specially reduce our development cycle. And with java you have lots of alternative PaaS: Google App Engine, Amazon Beanstalk, Heroku, Clodubees and much more. No lock in assured and good prices. Other reason to choose Java + Spring is that most SMBs and Big companies use Java as internal standard. If they are your customers or you are building a start up and an exit strategy for you is to get bought by one of these...then use java. And finally, regarding mongodb recommendations. Mongo is not transactional. You can mock it, but you can not mock a 3 phase commit, so if you are coordinating a transaction between more than one data store you just can't. In most cases I have been using a mix o SQL and No-SQL. Usually SQL gives as the transactionality and referential integrity. The No-SQL we use it for specific functionality and performance. Here some benchmarks regarding technology stacks...node.js is not even close to the top. Java servlet container with MySQL performs consistently well on I7 compared to the rest of frameworks (always top 3). When browsing the benchmarks be aware of the difference of platforms and fulls tacks: compare apples with apples! http://www.techempower.com/benchmarks/

REST API needs Security and Uniformity. As security is crucial, it is also tedious to code security checks. And offering a composable querying is another challenge. So we identified a new way to manage security rules as set of lambda expressions that will protect access to table & fields with simple rules. You only need to code your business logic, & security is taken care of automatically. In fact, your API code will be very small. It is a layer on top of Entity framework. http://entityrestsdk.codeplex.com

Google app engine - Able to handle any load at a reasonable price.  Scales beautifully and is pretty easy to develop in, especially for a rest service.  Many backends of iOS and android apps are built on app engine because of their speed and scalability.