You are going at it right by checking your memory. My concern is you lowered your voltage requirement but didn't adjust timings (loosen) to factor in the less voltage. It might be causing a stability issue.
I brought my clock down to 3.5 specifically to manage temps. IBT was getting 24/43/76 (hovering around 72C, under load) ambi/idle/load. P95 was probably doing about 10 deg cooler, so I didn't monitor it exactly. But it was in the 60's.
No, I didn't adjust the timings. So should I loosen the timings to this?
9 CAS Latency Time
Or add voltage to both qpi/vtt and dram-voltage and deal with a couple degrees higher heat?