Downvoting old answers based on score is a workaround for handling something the engine was never designed to handle: the passage of time. When version differences crop up, a weird thing happens: a single question with one or more answers morphs into multiple very similar questions (identical other than the version) posted in a single place. Answers to the different versions cannot be fairly ranked against each other, though that's exactly what sorting by score tries to do.
Sometimes, people have no choice but to use obsolete, deprecated or otherwise questionable technology. One common developer example: "I had to support IE6 until 2015 because my customer said so." Depriving these people of information they need—or, at least, an accurate measurement of the quality of that info—is unhelpful.
Making voting dependent on the score of an existing post confounds the meaning of a vote. The quality of a new answer doesn't change simply because a very high-scoring answer about a now-obsolete feature happened to be present, so its score shouldn't depend on that, either.
Original answerers shouldn't be punished with rep loss just because a feature they wrote about got deprecated years later.
This would have no effect on the problem for those weirdos who sort by age or activity instead of score.
At this point, you might be thinking "well, this is a flaw in the design of the engine". And you'd be right, but it's such a fundamental design element that there's no hope of changing it.
An alternate suggestion: in at least some cases, it might be worth editing posts where time has created ambiguity or problems. Edits could indicate that the information presented was absolutely right at one time, but applies to only version (whatever) of (product), and/or that it is bad to use old versions due to deprecation or security holes or compatibility, etc. And if there's a more modern answer present, it could point that out. Editing was always meant to be a big part of the ecosystem here, after all; a discussion of why people tend not to do it is beyond the scope of this answer.
(I originally wrote this post—and the edit to this post—as comments to PolyGeo's answer, but on second thought I decided it deserved a full answer.)