I doubt that they would form any codons at all, since the tRNAs bind to specific codons. They could bind tRNAs, I suppose, if they had the same H bond patterns on their edges as natural DNA bases. But then, why would one need non-naturally occurring bases?
If you wanted to create a whole genetic system—the base pairs, tRNAs that recognize the triplets, enzymes to charge the tRNAs with novel amino acids—I suppose that is possible as an academic exercise. Not sure how useful it would be.
Would there be any interesting 2o structures in the proteins from any new amino acids (etc. etc.)
All of which is to say something of a sensitivity analysis on the system as it exists, is the optimal or near optimal for forming animal/human proteins, compared to other, umm, ingredients (assuming the same general form DNA--> tRNA --> amino acid --> protein)?
...and would the current proteins which unzip and stabilize DNA (and correct mistranscriptions) work as well on the DNA with 8 base pairs?