I could be wrong, but IBT specifically stresses the CPU, I think. A blend test is testing the RAM as well (hence blend). I would double check timings and what not. Also as noted in the sticky just because you're getting x-x-x-x RAM at y frequency doesn't mean it's going to hold that for 100% sure when overclocking.
On the memory? You should use stock volts and adjust off of that with the latencies to other common XMP profiles. (going upwards in number of course). Unless you feel like overclocking the memory (upping the voltage) which is more often too much of a pain for what it's worth.