I modified the simulation I made up-thread for a couple of scenarios, out of curiosity.
bios64 and bios75 assume that NVRAM stuff only happens on the first draw, bios64d and bios75d assume that you get to do NVRAM stuff on both your first draw and mulligan (unlikely, but I can dream!). Odds of finding at least one copy of a 3x in your opening + mulligan + NVRAM:
45-card 0.5161
40-card 0.5609
bios64 0.7115
bios75 0.7397
andy 0.7472
bios64d 0.8312
bios75d 0.8655
7+5 for NVRAM is better for sure, but it’s not BANANAS.