I was expecting a considerable speedup from rewriting the whole rendering code as a C module and compiling it into the CircuitPython firmware, but a 50x improvement is much better than anticipated. I was afraid that I will have to limit myself to just a few sprites on the screen at once, and just one or two layers of tiles. But with this speed, there is no such problem. Of course, it will still slow down to around one frame per second if I fill the whole screen with animated sprites, but I think even ¼ of screen is already workable.