The sun.awt.image.PNGImageDecoder is slower than ImageIO's png decoder.

If the PNG image is 8-bit, and non-interlaced, then the code path it follows resembles:

```
while (col < width) {
    if(wPixels !=null) {
        <other code>
    } else switch(bitDepth) {
        case 8: bPixels[col+rowOffset] = rowByteBuffer[spos++];
            break;
    }
}
```

So we're calling `bPixels[j] = rowByteBuffer[k]` individually for width-many iterations. Using System.arraycopy or passing rowByteBuffer directly to the ImageConsumers would be noticeably faster