The CN642A is indeed not a direct driven printhead. Even the math simply doesn’t add up. A printhead of this age has to have hundreds of nozzles, yet there are slightly under 40 pads on the printhead. While I thought matrix for a while, the traces coming from the pads don’t support this theory, there are a few (around 8) really thick traces (2-4mm thick) and the rest is hair thin traces. Together with the amount of all traces, current theory is that there are shift registers somewhere in the printhead. There is 3 traces for every color and 5 sets of 3 for the black.
I myself do not see any problem with your setup, but please don’t use my word as guarantee that it is safe, I have no real experience with zener diodes. If you have a printer with a CN642A and a logic analyzer, it would be great if you were capable of getting a data dump from a printhead. It shouldn’t be too hard to establish if it is indeed a CN642A from the signals being sent to the printhead.
I like the idea of printing shirts with it. I hope that the printhead is capable of printing the kind of inks required for T-shirts. It shouldn’t be impossible to design something to print shirts once the printhead is hacked. One question, wouldn’t it be possible to simply hack an existing Inkjet printer to directly print on T-shirts? This should give better resolution and is very doable with current knowledge and tech.