Since commit 0d519164f29eb04208eac8bb7dd85443e2b282e1, qt_gl_read_framebuffer has been trying to smartly read textures as BGRA to avoid post pixel manipulation. On little-endian machines a byte-order of BGRA is identical to Qt's internal format of 32bit ARGB, but is completely wrong on big-endian machines where it becomes 32bit BGRA.

The former code was not optimized, but was appears to have been more correct.