Firstly, to answer your first question, QString is definitely not immutable. Directly from the documentation:

This operation is typically very fast (constant time), because QString preallocates extra space at the end of the string data so it can grow without reallocating the entire string each time.

I'm not a fan of the nested ternary statements in tool_unhex. Also, it does no error checking to make sure that whatever it is passed is actually a valid hex character. This is dangerous, as you use an unsigned short to store the shifted value of the character. If it's some character that's above F or f, then shifting and adding will potentially overflow and cause hard to track down bugs. For example, if there was a g that got in there somehow:

'g' - 'a' + 10;   // Equals 16
unsigned short w = (16 << 12) + (16 << 8) + (16 << 4) + 16;  // Overflow

I'd suggest changing it to the following:

#include <cctype>

// Converts the given (ASCII encoded) hex character to 
// an integer value 0 - 15.
// throw std::out_of_range if the given character is not
// a valid hex character.
int tool_unhex(char c)
{
    if(std::isdigit(c)) 
        return c - '0';
    else if(c >= 'A' && c <= 'F')
        return c - 'A' + 10;
    else if(c >= 'a' && c <= 'f')
        return c - 'a' + 10;
    throw std::out_of_range("Not a hex character");
}

If you're sure that the character passed in will always be in range, this should be documented somewhere. Also, it should be documented that this will ONLY work for ASCII and ASCII-compatible encodings.

Your unescape method has some confusing parts. The comments and the code don't match.

char* s = (char*)sOrg;

Why cast away the constness of sOrg? You don't modify s anyway, so just work directly on the (const) sOrg.

str = "";

If you're just going to clear the QString passed in straight away, why not just create and return it instead of taking it as a reference?

if ( *s == '%' ) {
    if(*(s+1) == 'u')
    {
        s++;
        if ( *++s != '\0' ) {
            w = (wchar_t) (tool_unhex( *s ) << 12);
        }
        if ( *++s != '\0' ) {
            w += (wchar_t) (tool_unhex( *s ) << 8);
        }
        if ( *++s != '\0' ) {
            w += (wchar_t) (tool_unhex( *s ) << 4);
        }
        if ( *++s != '\0' ) {
            w += (wchar_t) tool_unhex( *s );
        }

        str += QString::fromUtf16(&w, 1);
    }
}

This can certainly be simplified with a loop.

In the end, I came up with something like this:

QString unescape2QString(const char *sOrg)
{
    QString str;

    for ( ; *sOrg != '\0'; ++sOrg ) {
        if (*sOrg == '%' && *(sOrg+1) == 'u') {
        {
           unsigned short w = 0;
           ++sOrg;
            // Left shift value
            unsigned i = 12;
            while(*++sOrg != '\0' && i > 0) {
                w += (wchar_t) (tool_unhex(*sOrg) << i);
                i -= 4;
            }
            str += QString::fromUtf16(&w, 1);
        }
        else 
        {
            str += QString::fromAscii(s, 1);
        }
    }
    return str;
}  

This definitely needs some documentation as to what exactly it's doing, as it is absolutely not clear currently.