I do have some future-direction worries about php (the 6 dev process has had some ... quirks; like "let's just bin this and start over") and mysql (re: acquisition by oracle and ensuing staff departures, forked projects, etc.). But they're still fine for a proof of concept implementation. Not many architectures survive contact with real use unchanged, so having to buttress (e.g. op code or data caching) or swap out components entirely is not unheard of. Sometimes the best way to understand a system is to build it, so even if you do end up rebuilding it in whole or in part, at least you've learned the hard lessons about how everything hangs together to make it work right.
PHP and mysql are dead stupid easy to get started with. An mvc framework like symfony/kohana/zend/etc. can help speed dev time for your PoC, but most of them are performance hogs so you'll probably have to buttress that (APC, memcache, etc.) if you do end up having to scale. All this stuff's widely supported and retarded easy to deploy usually.
RoR is a bit more of a moving target at the moment. Which means docs can/do lag the bleeding edge quite a bit, standard OS distribution packages are frequently out of date (or just out right retarded e.g. debian's disabling of rubygem self update), and the best practices for deploying and scaling are in a state of flux. (I think currently the recommended way is phusion's passenger/mod_rails or whatever it is, for apache/nginx anyway. Or has the state of the art moved on again?)