Optimizing bilinear interpolation?

[superfury]

I've made a function to do some bilinear interpolation on some rendered VGA memory (just an array GPU.emu_screenbuffer[y][x], each item is a pixel to be shown).

//Divide!
#define DIVIDED(v,n) (byte)((double)v/(double)n)
#define AVGRGB(r,g,b,n) RGB(DIVIDED(r,n),DIVIDED(g,n),DIVIDED(b,n))

inline uint_32 fuse_pixelarea(word x1, word y1, word x2, word y2) //Fuse an area together into 1 pixel!
{
	uint_32 numpixels = (((x2-x1)+1)*((y2-y1)+1)); //Ammount of pixels!
	uint_32 r = 0, g = 0, b = 0;

	word y; //Current coordinates!
	for (y=y1;y<=y2;++y)
	{
		uint_32 *pixel = &GPU.emu_screenbuffer[y][x1]; //Start X to begin with!
		uint_32 *lastpixel = &GPU.emu_screenbuffer[y][x2]; //End X!
		for (;;) //Loop for each pixel!
		{
			uint_32 thepixel = *pixel; //Get the pixel!
			r += (thepixel&0xFF); //Red!
			thepixel >>= 8;
			g += (thepixel&0xFF); //Green!
			thepixel >>= 8;
			b += (thepixel&0xFF); //Blue!
			if (pixel==lastpixel) break; //Next row when needed!
			++pixel;
		}
	}
	
	return AVGRGB(r,g,b,numpixels); //RGB value of all pixels combined!
}

Anybody knows why this is being so slow? It takes about 5 seconds for the PSP to render 1 VGA frame this way (the CPU running at 333MHz). (This should at least be 0.125 seconds for full VGA speed, preferably lower).