I hear the recommended approach is to write your own module in the engine level code. But that means you need to compile the engine everytime you change your module. Possible, but you are losing a lot of iteration unless you are experienced in rolling in your own physics.

Nothing stop you from modifying the engine code, but the problem lies in the binding layer, not the engine perormance (as you see in the blog, the raw engine performance for a raycast is about the same as Unity). That's a bit trickier to fix without intimate knowledge of such bindings, and you'll likely break a lot of GDScripting support (so by extension, c# scripting) in the process.