> Another reason to use discrete logic rather than a 6522 is that some
> logic families can source and sink significantly more current than a 6522

Yeah, the NMOS ones were pretty lame. The CMOS ones OTOH could give you quite a lot. An experiment with one of my Rockwell 65c22's found that it could pull up and down to valid logic levels with 10 and 40mA. Pulling down, a short to Vcc gave 110mA per pin.

Yes, but they're specified much lower than that though. WDC parts, for example, are generally rated extremely conservatively, but it still depends on how far you're willing to push things, especially if the circuit is already operating at or near other parametric limits, like temperature or frequency. If the circuit is being built in large quantities, that might be a factor in this sort of decision as well.