Syntax highlighting is not context sensitive (at least not in this sense), unluckily. That's not how it works. It matches some control structures, such as comments, and it can distinguish balanced and unbalanced braces, but other than that it simply matches substrings.
So unluckily, I see no way how one could fix this, other than rewriting SciTE's syntax highlighter entirely and combining it with a parser that is superior to the one we currently use for code completion. In one word: no.
(of course removing all of the STL highlighting would be another option)