Diminishing upper limit on Rubik's Cube solutions - why so long?

The nodes in your tree you described are not unique. Sure you can fit 43 quintillion nodes in 25 levels, but because your tree allows for duplicates, you can't be sure that you got all unique positions of the cube.

For example consider the node reached by turning RL compared to the node reached by turning LR.

I think that's what you were getting at with your Rule #3: trying to delete obvious duplicates, but in that case each node no longer has 6 children, therefore you no longer can do $\lceil\log_b n\rceil$. For example how many children does the node reached by RRR have?

FWIW you don't need to define a one-direction-only metric to do this (unfortunately flawed) analysis. In regular half-turn metric, each node will have 18 children. So $\lceil\log_{18} n\rceil=\lceil15.6428\rceil=16$, which is way less than what we know is required, due to so many duplicate nodes being in our tree.

(Credit where credit is due: this is basically @Arthur's answer he gave in the comments.)