Using a PAL/GAL is kinda "period accurate" in a sense, as they would've been around at the time. In my case I had to compare 10-bit numbers and it was a huge pain in the arse trying to find a fast, easy way of doing that.
Perhaps the most important thing I learned during my initial experiments (other than how much piss you can take with timings and decoupling and have something still work!) was that Atmel 16V8's CAN be programmed with cheap TL866 programmers with no fuss whatsoever. I've lurked for long enough to know that Atmel SPLDs are treated with a lot of suspicion, and while the 22V10s don't work with the 866, the 16V8s definitely do and are extremely cheap for the functionality you get. They will now feature in all of my oldskool video controller designs.
The other possible advantage of using them is that assuming you end up using a socketed oscillator module, you could support a lot of other video timings/standards with one circuit design. Swap out the oscillator, adjust your sync and blanking timings, and change a VGA compatible signal into a 15KHz 50Hz one instead...
I'm currently waiting on some PCBs to turn up which should, in theory, give me a 64*48 display (512*384 resolution), with 16 colours, foreground and background colour per tile, and RAM-based character tile memory. If it actually works, and you haven't yet found a solution you're happy with, I'll gladly give you everything I have. At present it uses about 30 74HC chips, but I am not a professional design engineer so no doubt it can be optimised!