This function is about 8 times calles for every single page call, when caching this variable I was able to gain a small performance improvement from 20,512 µs to 630 µs profiled with xhprof

Surely, this is no gigantic gain but if we would do that for every function out there...