You are right; if we need about 1 second a year that is about 30 minutes in 2000 years. Lets agree to add a leap hour 2000 years from now - that will put us 30 minutes ahead so we can go 4000 years until we need to add another hour

Bad idea for things that use the sun to synchronize - solar panel servos, for example, or locator buoys that need a failover in case GPS satellites all decide to call it quits (There are such things for navies and commercial ships).

If some system is so disconnected that their direction/location can't be adjusted remotely every few years or so, then the current leap second data doesn't work for them anyways - they aren't scheduled far in advance, and that system won't get the info when a new leap second is announced.

By the way, do solar panel servos really have the proper sun-time set in them? I mean, if one panel is 200 feet west of another, is its clock set to be 2 seconds later?

Then those few super-delicate astronomical systems can sync up their own leap seconds.

Leap seconds are a terrible, idiotic thing because some people want to have their NTP servers lined up with sundials. There's no legitimate purpose for them in use for computer nor civil time.

Proposing a leap hour seems like the best way. That way we can nod to the people that feel the sun's position matters, while knowing that in thousands of years someone else can deal with the issue.

The sun moves 0.25 arcminutes in the sky every second, and has an apparent diameter of 30 arcminutes. Meaning that even after 50 years, if you (or your solar panels) look straight up at noon you'll still be pointed at the sun. So systems built to last for hundreds of years are the only situations where I could see it mattering.

Even less than that - it affects only systems that are both built to last for hundreds of years and have no maintenance or communications possible for hundreds of years and aren't smart enough to carry a bit of code to do an approximate adjustment themselves. Which should mean no systems at all.

The only one that comes to mind is Stonehenge.

Sun-tracking monoliths in general. We should really get around to building one of these near the seed vault at some point.

Keyword being "should".

Wouldn't it be more efficient, and less error prone, for solar panel servos to have a sunlight/brightness input feedback mechanism, rather than time-of-day based position?

How long is any one buoy in service? 5 years? 10? will it really matter on that scale?

Couldn't we just redefine the second by a tiny amount so it fits the day?

Unfortunately the day fluctuates, and is actually lengthening over time: http://en.wikipedia.org/wiki/Fluctuations_in_the_length_of_d...

We need at least one unit of time which is fixed and unchanging in order to do scientific calculations, and we have chosen the second to be that unit.

So, instead of adding or subtracting leap seconds, we adjust the speed of light (and every other physical constant that includes a time unit)?

Also, which day would become the new standard: the solar day or the siderial day?

That would be a really terrible approach. Time is the ultimate basis for practically every other quantity we know how to measure. The SI second definition is literally the last thing you'd want to mess with.

This (seconds being a universal unit of time regardless of planet lived on) is exactly the situation described in Vernor Vinge's (a CS professor) "A Deepness in the Sky".