Given this fact, is there anything else I should consider?
Why not use QImage::constBits()? If you have converted to 8-bit grey scale, then each character in that array will be a single pixel. It doesn't make a copy of the underlying data, so no overhead there.

If you write your own grey scale converter, you can operate on the original ARGB bits without creating a new QImage. Convert to grey scale and average in the same loop. I am sure you can find code for grey scale conversion online or in the QImage source code.