I think it's important that you mention explicitly that your patching method of code is not safe for multithreaded code...
If the patch size if > 8 bytes (which isn't true for your patched calls) or a function call site is located in a cross-cache-line memory address (on a 64 byte boundary, basically) and two threads try to execute the function, one of them might modify a part of the 5 byte patch, and the other might only see part of the patch, and attempt to execute an undefined instruction...

Interesting to see results of applying BenchmarkDotNet, including RyuJit results.