Given the code:
paste(c("a", "b"), collapase = "\n")
One might surmise that the user intended to type 'collapse', but instead wrote 'collapase' by accident. R executes this code without warning:
> paste(c("a", "b"), collapase = "\n")
 "a \n" "b \n"
although this is not the desired result. It might be worth:
1) Having paste warn when `...` receives named arguments, or
2) Have R CMD check warn if it sees calls to 'paste' of this form (ie, with explicit named formals not called 'collapse' or 'sep'.
I think (1) could be a bit controversial since in theory a user might write something like `do.call(paste, <...>)`, where the call to `paste` does receive named arguments, but still acts as the user expects. (2) may be a nice addition and presumedly would be a relatively small extension to the 'codetools' package.
I think that you're right that a warning from paste() would generate too much noise, so the codetools check is probably better. However, codetools is a recommended package, not precisely part of R, and I suspect its maintainer is quite busy.
What you might do is work out what the code for the test should look like, and submit a patch to him that either includes just that patch, or a way for users to supply their own checks to codetools. If you did the latter, then it would be feasible to add this check (and perhaps others) into the R CMD check code.