One factor is that different codons translate at different speeds, and these can affect how the protein folds.
Another is that base pairs in DNA may have functions or implications apart from their function as part of a codon in a gene.
Simply remapping a codon to a new amino acid and re-writing all the genes in a cell's DNA to avoid that codon will cause many things to "break" in a cell's function. Life is a messy, inter-connected system without the clean modularity we like to have in software.
1) My argument was simply that the limit on recodable codons is much higher than two, as it was claimed in the article. That's hard to argue with. I didn't say the result in every case would be neutral mutations.
2) The consequences of recoding the stop codons are also not neutral. For example, when Isaacs (eventually) did it, there were severe growth phenotypes in the resulting strain.
So I agree that "it's not that simple," which is frankly why I'm not hugely optimistic about this class of endeavors. But the point stands that there is no need in principle for a 4-base code.
So that leaves 43 codons, not 2.