I don't know what I'm doing wrong. Are my masks paths too complex with 49 points each? Are the shapes too different? Do I need still more correspondence points? Is the resolution still too high? I just don't know!
It's been awhile since I've used reshape but as I recall a general rule of thumb is to use as few correspondence points as possible to get the deformation looking right. I'd usually start with 2 or 3 and add more as necessary regardless of the assumed complexity of the masks.
Also, I'm pretty sure you shouldn't be using Mask 2 as a boundary mask. I think that'll wreak havoc if a correspondence point attempts to accommodate across the boundary (which is happening with the very mixed convex/concave shape of Mask 2).
I tried it with and without a boundary mask, didn't work. Started with less correspondence points, and it was worse. Added a bunch more, was less bad but still pretty bad, and while not making the problem worse I don't think adding yet more could've fixed it. Also, those correspondence points often didn't appear for some reason when I tried to add them, and the clickable area for moving them around was frustratingly small. So after much frustration I gave up on Reshape and went with Bezier Warps instead. I didn't get the result I wanted, but it'll have to do, what with the deadline.
Your masks were probably over complicated as well. There's really no need to bring the mask inside the folds of the shirt/clothing. Make the mask more convex overall and you'll probably get better results. It also looks like you may be using RotoBezier masks which isn't necessarily a problem, but I find much harder to work with generally.
I downloaded your problem image with the multitude of correspondence points and drew my own simplified masks with 5 correspondence points and had decent results in 2 minutes.