The difference in real-world use isn't nearly so pronounced in my experience. MicroPython is written in C, so whenever you're using modules implemented natively then you're only paying for a small boundary crossing.

For example, all of the machine API is typically very fast. I'd be surprised if you'd notice a significant improvement sending SPI buffers directly from C. But it's way more convenient from MicroPython - especially considering you can do it live at a REPL.

If you're spending a lot of time processing in MicroPython then yes, you'll notice a perf hit. But, if that's the case, carve out the perf-sensitive code and create a C module in MicroPython; it's straightforward.

It's akin to the way that CPython is considered slow but by using libraries like numpy or pandas the real-world performance is decent.