I was looking into this. The M5 performance cores can be scaled down to match efficiency cores in performance and power usage.

I believe they lower the clock speed, limit how much work is done in parallel on each core, and limit how aggressive the speculative execution is so less work is wasted.