I'm using my own set of CFLAGS when I build benchmark binaries of the server. In those I do not set the f(no)rtti switch at all. This build I refer to as "default" below. I also did two more builds, one with -frtti (referred to as "rtti") and one with -fno-rtti (referred to as "no-rtti")

I've done a sysbench OLTP run on all 3 builds. Results are in this Google Sheet

Conclusion: at least in this benchmark there is no significant impact from compiling with or without RTTI support. Performance difference between builds is within the normal fluctuations.