There seems to be a general cognitive bias in the programming world against maintenance work, which strikes me as a bit dumb because it's both harder and more important than starting new projects. But as long as it exists, I'm going to exploit it, and do high-visibility but useless projects like port Arc to JavaScript or rewrite Tetris in 3 hours instead of contributing patches to JQuery.
This is not unique to the programming world. Starting something and finishing it looks a lot more impressive almost evrywhere than making something that exis´ts better. Just think how much more impressive founding yet another non-profit looks on ones cv compared to doubling the projects of an existing one.
I suffer from the same cognitive biases as everyone else. Even though I know intellectually that getting patches into Rails now is probably harder than starting Rails was, I'm still more impressed by the guy who started Rails.
But if you contribute to Apache or Lighttpd, you're building a webserver. If you contribute to Python or create your own open-source language, you're building a compiler/interpreter. If you contribute to Lucene, you're building a search engine.
All of these teach you skills that can be very valuable if you start your own company. If you can take your server count down from 27 to 1, that's money that goes straight to your bottom line. If you can spend 3 hours writing a script that fixes a problem throughout your codebase instead of 3 months doing it manually, that's time that you can use to innovate instead. If you can index the contents of the web efficiently, you'll know how to index lots of other things efficiently, and there's bound to be a startup idea in there somewhere.
I don't think "writing some boring glue code for the frameworks our for-profit businesses use" is all that much of a resume boost anyway. Writing Mongrel or Factor or Rails, however...