Addition: They are also displayed incorrectly on special pages, such as Special:AllPages.

There is nothing we can do about the fact Unicode bidi algorithm is just an awful thing. Not as long we don't know in which languages titles are written in.

With the following CSS rule they at least show wrong in "a beatufiful way":
a { unicode-bidi: bidi-override; }

It's not so awful if LRM, dir="ltr", or other solutions can be used. But in this case LRM is not allowed and there's currently no way to know whether a title is RTL or LTR.

Maybe LRM can be allowed? And a big warning will be displayed if it is used in a title? And maybe only admins will be allowed to save a title with LRM?

I tried applying "a { unicode-bidi: bidi-override; }" using Firefox' DOM inspector and it became "(efiL ruoY fo emiT) ecnaddiR dooG".

Yep, like I said it will display wrong with that rule. LRM and other characters are not allowed in titles because they break lots of things (including users typing the titles manually, possibly search..)

To solve the manual typing problem, redirects can be created. They can even be created automatically.

see also: https://gerrit.wikimedia.org/r/#/c/44766/

I can reproduce the problem with Chrome on https://he.wikipedia.org/wiki/%D7%A7%D7%98%D7%92%D7%95%D7%A8%D7%99%D7%94:%D7%A1%D7%99%D7%A0%D7%92%D7%9C%D7%99%D7%9D. With the following CSS rule I can eliminate this problem:

a {
	unicode-bidi: plaintext;
}

In T30411#2959977, @Fomafix wrote:

I can reproduce the problem with Chrome on https://he.wikipedia.org/wiki/%D7%A7%D7%98%D7%92%D7%95%D7%A8%D7%99%D7%94:%D7%A1%D7%99%D7%A0%D7%92%D7%9C%D7%99%D7%9D. With the following CSS rule I can eliminate this problem:

a {
	unicode-bidi: plaintext;
}

Nice idea! This looks good for page titles ending with symbols(?!). However, it seems to have unwanted side effect sometimes - e.g in the above category notice "Umbrella (ריהאנה)" - where the word in parenthesis should be to the left of the song name and becomes to the right.