I don't think my benchmark is misleading. I set out to show that in at least one case prepare() CAN be faster than prepare_cached(), to provide a counter example, as it were. The benchmark that I posted shows that in some cases prepare() CAN be faster than prepare_cached() nothing more, nothing less.
Now, prepare_cached()is normally faster and I am not saying
that it is not, but there a point where the overhead of calculating the hash and doing the hash lookup takes more CPU time than it does for DBD::mysql to walk the string looking for placeholders.