I can't speak to what's going on under the hood regarding the function calls vs. function arguments, but this email from Prof. Ripley from 2008 may shed some light (excerpt below):
R passes around, prints and plots UTF-8 character data pretty well, but it translates to the native encoding for almost all character-level manipulations (and not just on Windows). ?Encoding spells out the exceptions [...]
The reason R does this translation (on Windows at least) is mentioned in the documentation that the OP linked to:
Windows has no UTF-8 locales, but rather expects to work with UCS-2 strings. R (being written in standard C) would not work internally with UCS-2 without extensive changes.
The R documentation for ?Quotes explains how you can sometimes use out-of-locale characters anyway (emphasis added):
Identifiers consist of a sequence of letters, digits, the period (.) and the underscore. They must not start with a digit nor underscore, nor with a period followed by a digit. Reserved words are not valid identifiers.
The definition of a letter depends on the current locale, but only ASCII digits are considered to be digits.
Such identifiers are also known as syntactic names and may be used directly in R code. Almost always, other names can be used provided they are quoted. The preferred quote is the backtick (`), and deparse will normally use it, but under many circumstances single or double quotes can be used (as a character constant will often be converted to a name). One place where backticks may be essential is to delimit variable names in formulae: see formula.
There is another way to get at such characters, which is using the unicode escape sequence (like \u0394 for Δ). This is usually a bad idea if you're using that character for anything other than text on a plot (i.e., don't do this for variable or function names; cf. this quote from the R 2.7 release notes, when much of the current UTF-8 support was added):
If a string presented to the parser contains a \uxxxx escape invalid in the current locale, the string is recorded in UTF-8 with the encoding declared. This is likely to throw an error if it is used later in the session, but it can be printed, and used for e.g. plotting on the windows() device. So "\u03b2" gives a Greek small beta and "\u2642" a 'male sign'. Such strings will be printed as e.g. <U+2642> except in the Rgui console (see below).
I think this addresses most of your questions, though I don't know why there is a difference between the function name and function argument examples you gave; hopefully someone more knowledgable can chime in on that. FYI, on Linux all of these different ways of assigning and calling a function work without error (because the system locale is UTF-8, so no translation need occur):
Δ <- function(a,b) (a-b)/a # no error
`Δ` <- function(a,b) (a-b)/a # no error
"Δ" <- function(a,b) (a-b)/a # no error
"\u0394" <- function(a,b) (a-b)/a # no error
Δ(1:5, 9:13) # -8.00 -4.00 -2.67 -2.00 -1.60
`Δ`(1:5, 9:13) # same
"Δ"(1:5, 9:13) # same
"\u0394"(1:5, 9:13) # same
sessionInfo()
# R version 3.1.2 (2014-10-31)
# Platform: x86_64-pc-linux-gnu (64-bit)
# locale:
# LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C LC_TIME=en_US.UTF-8
# LC_COLLATE=en_US.UTF-8 LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
# LC_PAPER=en_US.UTF-8 LC_NAME=C LC_ADDRESS=C
# LC_TELEPHONE=C LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
# attached base packages:
# stats graphics grDevices utils datasets methods base
