Your experiment is based on a false premise. A figure 8 mic picks up front and back 180 out of phase with the other, physically creating a perfect rejection zone to the sides. Just check out a figure 8 polar pattern.
2 cardiods back-to-back would need to be connected or mixed together with one out of phase with the other to emulate the figure 8 with the side rejection zone. Connecting or mixing 2 back-to-back cardiods in phase gives you a virtual omni, which is why it doesn't work in an M-S setup. Adding an inverted signal of the already-combined in-phase cardiods as you've done, is not the same - you just made 2 emulated omnis out of phase with each other, not a figure 8.
Here's the proper process - combine 2 cardiods back-to-back with the one pointing right flipped out-of-phase (if they're identical dynamics, you can get away with a y-cable with the right FE XLR pins 2&3 wired out of phase), designate that combined signal "L in-phase", pan L on mixer. Phase flip a duplicate of that combined L in-phase signal to generate a R out-of-phase, pan R on mixer. Add front facing cardiod panned center.
Even with the back-to-back cardiods combined out of phase with each other, because of the necessary physical spacing, they can never achieve as perfect a side signal rejection of a figure 8.
Further, many newer (I'm old) figure 8 mics use 2 capsules, each with its own diaphragm, and again due to physical spacing, though better than 2 out-of-phase cardiods, those still cannot achieve as perfect a side rejection of a ribbon figure 8, for which both front and back use the same diaphragm.
It's fine to emulate as best you can, and I sure have jerry-rigged setups when needed, but a ribbon figure 8 in an M-S setup is always the best way to go. Next choice, a 2-capsule figure 8.