As an owner of both the M and the 5D3, I share my thoughts:
I would guess that huge increase of the popularity of the M is mostly because of much more youtube videos about that: actually I have also found ML because an EOS M video just has popped up one year ago. And I have learnt only later that you can do similar things with other Canon cameras.
Price: as far as I understand, when all the ML development had started, the price difference between the M and the 5D3 was much higher, but now with the popularity of the M (and the drop of the 5D3 prices due to the advent of the R series), it is not that much of a difference any more, I agree. But it is still cheaper, so if someone starts to do videography, might feel safer to start with a more budget-friendly option.
Main advantages of the 5D3: to me it is a huge improvement over the M that it has a native 1920x1080 resolution with native 3x3 binning: the image is gorgeous even in the plain 1080p mode, and can be used up to 60 fps. (I personally don't really use other modes on the 5D3, 1080p is sharp enough to me on any monitor in a normal viewing distance. I really enjoy the manageable file sizes and quick editing.) Moreover, it is one of the best stills camera on the market: reliable autofocus and very good pictures straight out of the camera even in low light. For professional video and photo work I would definitely grab this one.
Main advantages of the EOS M: the best thing about it its small size and weight, with the 15-45mm kit lens I can literally fit into the pocket of my jacket, so I take this one if I go for a small excursion or meet friends. And even on a festival I can work quite unnoticed with the M. Another plus is the adaptability of the M mount, I sometimes use it with smaller EF-M lenses, or with EF lenses with the Viltrox speedbooster. The only downside for me is the line skipping in the 1080p mode, but the artefacts can be avoided using a shallow depth of field and/or soft lenses. For very detailed scenes using the anamorphic and the high resolution modes are the way to go, at the price of higher file sizes. And who knows, at some point we might get digital 3x3 binning, where the three lines are added before being written out to the card, to make a super1080p mode. After all, so many developments have happened that noone thought possible before!
So both have their place in my workflow and it is great that Danne is developing for both so I can use ML builds that have pretty similar functionality.