Does anyone know what percentage of 7-piece endgames are draws?

What everyone appears to miss is that castling in tablebases only figures prominently in endgame studies. You cannot consult them during a game and if you were to study them as part of your endgame preparation you wouldn't care about castling rights. The pay-off is simply too low.

Now the fun thing is that the effect of the castling right is mainly visible in connection with the 50-move rule. Castling right may speed up the mating process by a few moves which makes a difference when a 50-move draw is in sight. However, in endgame studies the 50-move rule does not exist which makes the whole discussion academic. Much ado about nothin!

Though, .... well I won't tell you because it's retro and retro guys always make trouble

 Were the endgame tablebases made just for 'endgame studies' or people doing such?
I would have thought they were made for various reasons including reasons far beyond those two reasons.
There's the relevance of the word 'solved'.
In most 7-piece positions in chess - castling would be irreversibly illegal.

The great majority of such positions - far over 99% of them 
But that doesn't mean one can then make a valid claim that 7-piece is 'solved'.
Disqualifying castling rights immediately 'taints' the project.
The availability of castling in a position - makes it unique and different from positions where its not available.  

Apparently the reason they could have 'en passant' still factored in but not castling - is that the en passant option doesn't persist. 
It lasts one ply deep. 

But with castling - anytime you have either (or both) King on its original square along with either of its rooks (or both) on its original square then the analysis has to change even if its temporarily illegal because of blocking pieces or checks or in check or the K-traverse square in check.

The tablebases - among other things - were apparently built with an idea of creating more tablebases with more pieces in them - and using the simpler tablebases to help solve the 'higher' ones.  (where that's possible - it isn't always so - and there's issues of how the higher and lower tablebases are 'communicated')
Anyway - it seemed to me at first - that generating the higher tablebases  looked quite straightforward.
Things aren't always what they seem to be.  Though.

Tempting to think - you could create the next higher tablebase by simply adding a piece - or by 'reverse legal capture'.
But neither of those follows anything like that simply.
If you've got Kings and/or other pieces on the 48 pawn squares - that's going to affect where pawns can be or go to.  Or be placed.   
Which in turn will affect calculations as to available squares. 

If the Kings or other pieces are on the back ranks - then there'll be different math.
Similiar with the bishops.  Are pieces placed on the 32 squares a particular bishop can be placed on?
All these factors mean that more groupings are needed in the more primary tablebases.
Otherwise there might not be proper 'compatibility' with the higher ones.
If that isn't so or doesn't matter then why would castling be such a pain for the tablebasing?
Does it become easier to understand when its considered that 8 pieces hasn't been 'solved'?
There aren't just computer speed issues here.  There's explosions of groupings - and the math changes.
As pieces are added - if its attempted to just have the computer take care of those things by 'brute force' then mathematically - the computer is being assigned more and more extra (badly 'created') work (that multiplies on itself and grows exponentially) while it doesn't have the time to do the true task as it is !
 

So there's at least two discussions here:
7 piece positions draw percentages discussed computer-free and tablebase-free.
Or -
Factor in the tablebases as central to discussion - and as a peripheral subject too.

You will find indeed find that tablebases are overwhelmingly consulted by puzzle makers and puzzle solvers. 

Whatever interest theorists have in them happens behind the doors of scientific laboratories and aims to conquer the world. Science always aims for just that.

In the chess domain you only need to watch the history of the 2N vs P ending to realize that humans will never master the advanced endgame wins found by the tablebase miners. 2N vs P is only the intro chapter!  Do you think anybody will ever study and conquer a R+B vs R+B+N endgame by interrogating a tablebase? All they may learn from it is that such endgames commonly do win and for the rest they are on their own in a game. And whether they knew the tablebase outcome or not they would still plough on for dozens of moves hoping for an opponents mistake - now, 10 years ago, 50 years ago. The tablebase changes nothing to their approach of the endgame though their foreknowledge now is on a higher level than 50 years ago.

But there may come a time when chess (and many other games) will be upgraded to permit the use of certain digital tools while playing it. Probably chess will first have been made more intricate before that happens!

Regarding post #25 - and some good points made in it -
it occurred to me a while back - that while computers might 'solve' various unsolved positions and prove they're a win or a forced draw in many cases - the 'explanations' or 'proofs' of same might be both so deep and so numerous - that a human wouldn't have time to follow such anyway.
In another forum I saw a position the computer took over 500 moves to force win material in and from there win the game.  
How many people are going to 'follow' the 500 move 'proof' ??
There are under 32 million seconds in a year.  
Means there are under 4 billion seconds in 100 years.  
So nobody has 10 billion seconds - that's ten to the tenth seconds.
Computers don't have that kind of time either.  

For miniature or tablebase positions (7 units or fewer), castling is even more popular in directmate problems (Mate-in-N) than endgame studies. There are thousands of such directmates that involve castling in problem databases. It's a pity that tablebases can't solve such problems. See example below.

I think that eventually - the tablebases will be upgraded to factor in castling rights.
But how many years from now?
It appears the 8-piece tablebase might be completed in a few years or less.
But if they're using the same 7-piece tablebases to go with it -
that means that again - that new tablebase will be somewhat tainted.
Maybe by the time they have 9-piece completed - that'll be fixed.
Will 9-piece be completed - either way - during our lifetimes?
Each adding of a piece or pawn greatly increases the difficulty. 
And in multiple ways.  But there would be increases in the increases.  Why not?
The computers are not improving every year in power - at the rate in which difficulty increases from adding a piece.
Maybe not even every ten years.  
The power of the computer isn't just related to speed - there could be problems in programming algorithms.  And with the math for them.
Why do I think so?  
Two reasons - the math issues of adding a piece - and the fact they're having trouble with castling rights.
The factor of difficulty would appear to increase too.  

In discussions about 'solving chess' there's much about 'weakly solving'.
For example - taking the square root of the number of positions with a certain number of pieces.
That could be as contrived or artificial as trying to insist that chess is a draw if nobody makes a mistake.
That may be useful as an operating idea - but doesn't mean its true.