I am starting to wonder if it would just be easier to take the damn gc3355's off the boards, create a new pcb that can handle the power needed for SHA and start a password cracking service

I haven't checked, but I'm not sure they are that efficient for sha256. The dedicated sha256 chips would probably work better if they could crack passwords. Not sure they can crack passwords, even if the algorithm is the same.
Undervolting the GC3355 for scrypt will still make it usable for some time, but it's clear that ROI will be hard if you haven't hit it yet. As usual, new coins, instamining, luck... Could help.
The scrypt asic market is currently moving too fast, and prices are dropping in the same way. We should have something more stable when 28nm will be widely available. prices for hardware will level around 6-7$/MH and after that, it will only slowly decrease.
It'll be the best time to buy bare chips and build custom hardware, since the prices won't be cut by 10 or 15% every week.