There might be bug in the cost calculation in the browser (it might cache an old value ... I'll keep an eye out).

How's the higher cost experiment looking so far?

Qualitatively at least it looks pretty good. Recent is way less noisy afaict