Being able to easily link C++ modules with your Node program is a powerful new paradigm. It may be tempting, then, to exuberantly begin writing C++ add-ons for every identifiable segment of your programs. While this might be a productive way to learn, it is not necessarily the best idea in the long run. While it is certainly true that in general, compiled C++ will run more quickly than JavaScript code, remember that V8 is ultimately using another type of compilation on the JavaScript code it is running. JavaScript running within V8 runs very efficiently. As well, we don't want to lose the simplicity of organization and predictable single-threaded runtime of JavaScript when designing complex interactions within a high-concurrency environment. Remember: Node came into being partly as an attempt to save the developer from working with threads and related complexities when performing I/O. As such, try and keep these two rules in mind:
Is a C++ module actually going to run more quickly?: The answer isn't always yes. The extra step of jumping into a different execution context and then back into V8 is wasteful if your add-on is only returning a static string. Felix Geisendorfer's talk describing his work with building fast MySQL bindings provides some insight into these decisions: http://www.youtube.com/watch?v=Kdwwvps4J9A.
How does splitting up your codebase affect maintainability?: While it would be hard for any developer to suggest using less-efficient code, sometimes a negligible performance gain does not overcome an increase in complexity that can lead to harder-to-find bugs or difficulties with sharing or otherwise managing your codebase.
Node merged a beautiful JavaScript API with an enormously powerful and easily extensible application stack. Given the ability to integrate C++ into your applications, there is no reason to exclude Node from the list of technologies to consider for your next project.