When you want to measure timing differences between two different perl interpreters you want to use an independent timing facility. The time program works perfectly well for this task (and my system was under very low load, and I repeated the timings several times with consistent results). Yes my timings also measure startup costs, but that is insignificant as an absolute factor and even less significant as a relative factor. Just to ease your mind vis-a-vis startup costs, the times to fire up each interpreter with strict and -w are roughly 0.011 and 0.015 seconds respectively. These times are almost certainly in the realm of measurement error and we should only use them as a basis to conclude that startup costs are an insignificant factor in the 8 second difference in runtime between the two interpreters.

but 50% sounds too horrendous for any developer to accept.

I am not claiming 5.8.0 with threads is 50% slower across the board, just that it was 50% slower running one particular program. Please do run some of your own tests and post your results here.