The point being raised is that dynamic image grids don't actually make for a good UX. They might look more visually interesting at a superficial glance, but when you're actually using the interface to browse images, predictability wins out. Even having mixed-orientation images, where there is some degree of extra whitespace between images, does not change this. It is way easier to digest the content when your eyes can reliably scan one line at a time without having to bounce around everywhere to track the flow of the dynamic grid.
What is it with commenters in this thread and wanting to "reliably scan one line at a time?" When users use image galleries, they generally do jump around because they're looking at all the options on screen all at once. The eyes absorb everything and then they pinpoint what looks good. I've never seen or heard anyone go line by line in an image gallery or a newspaper layout and doing so I'd find to be highly abnormal to average users.