What's the reasoning behind that general recommendation? From your description, it appears that splitting on 'foo' and /foo/ (or even on $bar, if $bar eq 'foo') would be equivalent, so why prefer the pattern version?

Because the documentation says

       split /PATTERN/,EXPR,LIMIT
       split /PATTERN/,EXPR
       split /PATTERN/
       split
[download]

In fact splitting on 'foo' and /foo/ are (currently) equivalent because perl converts the string to a regex:

perl -MO=Deparse -ne 'split "foo"'
LINE: while (defined($_ = <ARGV>)) {
    split(/foo/, $_, 0);
}
-e syntax OK
[download]

Who knows? I doubt it, but the docs do not specify what should happen with generic strings other than ' ' and maybe one day special behaviour may be implemented for them along the lines of what is currently available for ' '.

(My current practice is to use 'foo' on the principle of never using regexes where a simple comparison is sufficient. On the surface, at least, it seems that splitting on /foo/ is the same manner of overkill as using if $bar =~ /^foo$/ instead of if $bar eq 'foo'.)

Your principle is based on an assumption that is simply not true for as we've seen before, your string is converted to a regex anyway. And OTOH matches like /foo/ are internally optimized to basically do an index. So no overkill. If you don't trust me, just trust B::Concise:

$ perl -MO=Concise -pe '$_=split "foo"' >1.txt
-e syntax OK
$ perl -MO=Concise -pe '$_=split /foo/' >2.txt
-e syntax OK
$ diff [12].txt
20c20
< 6                          </> pushre(/"foo"/) s*/64 ->7
---
> 6                          </> pushre(/"foo"/) s/64 ->7
[download]

Notice that the only difference is an asterisk, which means "Do something weird for this op".

I'm not saying that I don't agree with you on the $bar =~ /^foo$/ vs. $bar eq 'foo' issue. Because I do agree. (Also because they're not strictly equivalent:

$ perl -le 'print "bar\n" =~ /^bar$/ ? "ok" : "not ok"'
ok
[download]