Using the right tool for the job comes to mind here, but half the time you can only see this with hindsight. And the other half the time, I'd guess, PHP is the only tool available. Unless cheap webhosts offer Python, Ruby and Node.js integration to complement the usual feature-set.
If you have more tools in the box, then you can start to consider whether to go full-stack with PHP (I'd no longer consider this route a viable option), or whether to modularise your project. Maybe for a lot of things that's too much and a PHP framework that does it all is totally fine. No point doing something just because you can, but you have the flexibility to think outside of the box when refactoring.
This is exactly what I did when it came to generating an image and manipulating the output. In PHP it was convoluted, buggy, impossible to understand the library it depended upon, and offered no flexibility. It took months of procrastination and avoidance before we had to get the feature working, so I ditched the lot and re-did it in Python, using Cyclone and a couple of lightweight libs. It took a day and a half to implement, with no knowledge beforehand, and it does absolutely everything we wanted (without any hacks).
It was at that point I realised that it was much more beneficial to look beyond the one-size-fits-all solution of a massive framework, which may be appropriate to some, but just as often isn't. And as has been said - very well indeed I must add - I don't want to be a [framework]-developer who knows not so much the language, but an individual or group's abstraction of it; I want to be a developer who knows what to do when that framework doesn't fit.