HN completely ignores the class of computer users who only need Safari, Mail and some variation of Word, and then live in these apps for years without any major upgrade. In my experience, these people don't even feel entitled to any upgrade and they don't add any complexity to any testing matrix. All they need are baseline security updates, and existing services better don't stop without a plausible reason (see MobileMe). I'm not thinking of MegaCorp here, but rather my own and friends' close relatives.
The machine that got excluded from 10.8 after only 4 years was not a rev-A MacBook Air in the hands of some early adopters but a Mac Mini, the most traditional Mac money can buy. That really leaves little hope that all my "slow adopter" relatives will ever leave the Windowsverse.
You can't just add programmers to mate cutting edge OS kernel features (ie leveraging recent videocards) with old hardware. None of us can speak authoritatively on the case in point, but it seems painfully optimistic to think you can pull this off without introducing a lot of complexity. And then, what happens if the 10.8 team tweaks the kernel during a point update? Yep, better double or triple the testing team.
"Oh, it looks like it'll take some work to get 10.8 working with all of our old AMD chipsets - either we delay release until fall (after Windows 8, unfortunately) or we put some of our "A" engineers on it."
Complexity is the keyword. Complexity and mediocrity go hand in hand. Backward compatibility is a noble goal in many ways, but every generation back of software/hardware you try to maintain exponentially increases the complexity you deal with instead of putting a better new product in user's hands.