FWIW, here's how URW dealt with it in its version of Cooper Black (essentially, putting the bar on the ascender only to the right).



<http://www.myfonts.com/fonts/urw/cooper/black-d/opentype-ps/260370/charmap.html?vid=260370&cpselect=uni3>

I don't think the URW solution is very convincing. In such an extreme case, I would a) remove the top serif and straighten that side of the stem, and b) lower the top of the bowl as much as possible without distorting the form too much, in order to create space for the crossbar. Something like this (hacked in Photoshop, so not very tidy):



It is more important in designing diacritic letters for languages to preserve the particular identity of those letters than it is to preserve the particular details of a serif. This means being aware of the primary distinguishing features of the letter, in this case a bar that crosses through the ascender, and coming up with a way to preserve those features while preserving as much as possible of the characteristic details of the typeface design.

interesting...
i had read in Typo 10 that "In extra bold characters, the part of the slash inside of the character may be omitted." So i wasn't particularly shocked to see the URW example (I just hadn't seen anything like it before). I personally think John's is a better solution; but not being a native reader of any language that uses this character, i don't know which is the "preferred" design.

Obviously, a different letter in a different culture, but Gunnlaugur Briem suggests that for the lower case eth, shortening or omitting the bar can be done as a last resort when designing heavy versions of the glyph. See the last paragraph at: <http://briem.ismennt.is/2/2.1a/2.1.4.eth.htm>.

Whether or not this also applies to the Croatian dyet is an interesting question. I do like John Hudson's mock-up a better solution than the URW sample, though.

I suspect the comment in Type 10 is based on examination on what has been done in some fonts, rather than consideration of whether these solutions are good. As a last resort, I think the URW solution is acce