original submission:

> they should think about the order in which they specify the
> characters, as only the last one is likely to end up visible.

An exception is (at least some) space-like characters.

$ echo "\o'abcd'efg" | nroff | cat -s
defg

$ echo "\o'abc\0'efg" | nroff | cat -s
cefg

$ echo "\o'abc\ 'efg" | nroff | cat -s
cefg

Groff and Heirloom's nroff produce identical results with those inputs.  Where they diverge is when \o is given an ordinary space:

$ echo "\o'abc 'efg" | nroff | cat -s
troff:<standard input>:1: error: expected ordinary or special character, got a space
cefg

Heirloom nroff's output is the same, but it doesn't throw the error.  If that holds of ancestor nroffs as well, groff objecting to this formerly permitted construction should be documented in groff_diff(7).

original submission:

> (And maybe (3) that if they want to use \o for accented
> characters, there are more robust ways to achieve that.)

Nowadays, most of groff's Texinfo examples try to exhibit groff best practices, but the manual does have an example (polished by bug #63002) that uses \o to produce an accented character.

@Example
A caf\o
e\(aa
in Paris
    @result{} A café in Paris
@endExample

Despite this not being a good way in general to place diacritics on characters (it's particularly ill suited to the letter i, which must shed its dot when it acquires a diacritic), the advantage to using it in an example that many readers will view on a terminal is that a terminal can give meaningful output for it (albeit not through the \o mechanism but by using the appropriate precomposed character).  This would not be the case with an example that actually requires the overstrike to do its job, such as this style-nerdy one:

\[lq]I'm torn between US and UK punctuation styles\o
,\[rq]
she lamented.

However, taking context into account: this example isn't even trying to illustrate the \o escape, but groff's ability to use newlines as escape-sequence delimiters.  So maybe a different escape altogether would do this job without the side effect of choosing between best-practice groff code and terminal-displayable output (notwithstanding the fact that using newlines as delimiters is certainly not a best practice itself.  But at least that is discouraged in the very next sentence.  There's no comment on the overstriking-to-apply-diacritics aspect of the example--and such a warning would be out of place here, since the use of \o is not the point of this section).

I don't have a suggestion offhand of a better escape for this; I'd need to peruse the list of escapes that accept newlines as delimiters and choose one that can produce output that would render well on a terminal.

Lack of true overstriking will also afflict HTML output.