Your code calculates (dir - dir * dot(dir, norm)) * 2, whereas you should be calculating dir - dir * dot(dir, norm) * 2, that is:
Vector3 ReflectionDirection = newRay.direction.nor().cpy().sub(surfNorm.scl(newRay.direction.cpy().dot(surfNorm)).scl(2.0f));
Or just:
Vector3 ReflectionDirection = newRay.direction.nor().cpy().sub(surfNorm.scl(2 * newRay.direction.cpy().dot(surfNorm)));
This is incidentally the main reason why I strongly dislike the lack of (reasonable) operator overloading in languages, it makes it impossible to write such things in a sane way (and, no, separating this simple equation into three or four temporary variables is not sane either). But anyway this is where the pinching comes from.
(actually it's pinching not so much because the reflection is wrong but really because of your error the resulting reflected vector is not unit length, which wreaks havoc on the rest of the ray tracing code something fierce; so a good thing to do whenever stuff looks really weird is to look at your directions vectors and check if they are actually unit length; if they aren't, you've screwed up somewhere)