I've used CakePHP for a lot of projects now. There's no comparison to Ruby on Rails. Rails is smartly integrated, well-designed, and sits on top of an elegant language. CakePHP is an admitted ripoff of Rails that sits on a popular but terrible language. It copies Rails - almost randomly - without copying any of the stuff that makes Rails great. And PHP is a language plagued by indecision, delays, and paradigms from the last decade.

What makes Rails great?

• An ActiveRecord implementation that isn't brain-dead. The Object Relational Model is, you know, actually object-oriented.
• Lazy-loading of records that allows as-needed fetching of model associations instead of a dumb and unintuitive $recursive variable.
• Actual documentation that doesn't suck (not that it didn't suck in the early years). Cake's has been getting better lately, but it's still got a long way to go.
• Massive community of great libraries and plugins.

What makes CakePHP awful?

• Arbitrary and seemingly random nesting of HTTP input parameters. It gets to the point where I sometimes access $_GET and $_POST directly, and just carefully sanitize the data myself.
• Default routing does not use REST, and is difficult (but not impossible) to coerce into anything resembling REST.
• Model instances are represented as PHP arrays instead of objects. So it's like a struct, except uglier, and you can basically set a few class variables on the model, then call a class method, which simulates an object method. Nice. The CakePHP core team's position is: this is a deliberate design decision so users cannot call instance methods in the view. So not only is the decision based on stupid reasoning, it's not even a technical one.
• Ridiculous schema manager that has noble intentions, but tries to be too clever. Instead of making you write your own schema changes like Rails does, it tries to infer the changes after the fact (by diffing the actual schema) - with disastrous results. Let's say you rename a column and want to use the cake console to capture that change in the schema manager. Now let's look at the generated schema change that was automatically inferred. Instead of renaming, it will drop the column and make a new one, which means - yes - all your old column data is now gone. This is because there's no way to tell with certainty the difference between a rename and a drop/add after it has already happened. But it matters. What they're trying to do can't be done. Instead they try to do it anyway and make users deal with the data loss.

Cake doesn't even try to hide the fact that its cloning Rails in PHP, but it's at a disadvantage from the start because PHP's syntax is just so much worse. And they use arrays to represent objects. And PHP's array syntax is terrible. Unbelievably terrible. Which makes the decision to go all-in on PHP arrays baffling. I spend the majority of my time in Cake trying to manipulate arrays-of-arrays-of-arrays-of-arrays that are supposed to represent some kind of shoddy struct, then mangling with model associations that require about 5 nested arrays to define. Add to this the fact that CakePHP tries to "innovate" its way around some of its shortcomings in a less-than-brilliant way (like, um, $recursive) and you get a steaming pile of garbage that's only marginally better than just coding a PHP project from scratch.

Rails's smart design has come at a cost. In the past, they haven't been shy about breaking backwards compatibility, sometimes in a big way, to cut out cancerous ideas before they take hold. It's a community that acknowledges mistakes and embraces their eradication. And Ruby is an elegant language with strong design and is under active development. PHP on the other hand has languished for years, most notably with a critical security bug called $register_globals. On top of that, they've been slow to introduce object support, type-checking, and sometimes make terribly confusing decisions (like the decision to call a feature overloading that has absolutely nothing to do with method overloading in the traditionally accepted sense of the word).

To be fair, here's what I don't like about Rails (and Ruby):

• Massive list of reserved variable names, which of course will throw a cryptic and seemingly unrelated error if you use them.
• Concise syntax that sometimes resembles perl. Yes, perl.
• Bizarre decision to use "end" to close blocks instead of something sensible, like um, anything else. More typing for no good reason. The cherry on top that makes this not OK is that while its illegal to define a variable named "end," its fine to define a function named "end."