Wouldn't it be a way to increase the output for witnesses, including witness-decay votes? Or is there a big argument against it? Usually, witnesses are a great group of people competing for a top spot, and in thus trying to provide as much value for Steem as possible.

I don't know what difference you think witness decay would make in the short term. What share of the witness votes, stake-weighted, are stale in the sense that voter is inactive or doesn't care? Because if the voter is active and does care they will just refresh their votes (almost certainly the case of the larger ones, with a bot, meaning no more human attention will apply than now). IMO it is very low and won't change things much if at all (the largest stake-weighted voters are all active as far as I know).

I do support vote expiration (I don't see a feasible way to implement decay in the consensus logic) but it is more of a longer-term theoretical concern to me (dead voters, etc.) and not something that is going to make a real difference now. In that sense I'd suggest we focus our efforts and resources (limited though they are, relative to other projects, as per other discussion in this thread) on things that matter.