Presuming you're a great engineer that dislikes PHP, you'd take a crappy language and build hacks on top of it? Your great engineers would probably have better things to do with their time. Artificial constraints are great for creativity; for engineering, they're just time-wasting and your good engineers will go elsewhere.
(unless you're the single exception where your business is just that compelling...)
PHP isn't even a good compilation target. There's no native FFI (you can't write pure PHP code that makes arbitrary syscalls), the extension interface exposes extremely brittle guts of the interpreter, it performs very poorly (compared to, e.g., JVM bytecode), and calling user-defined functions to work around the stupid misbehavior of the builtins (e.g., == which isn't even an equivalence relation) is only going to make its performance worse.
Forced to use PHP? And it's crappy and slow to build stuff in? So write some tools that compile to PHP.
What I'm saying is that great programmers work with the limitations and build the tools they need to make the limitations moot.