Rectangular packing works as you expected: it tries to pack in the smallest size possible even if the max size allowed is much larger. It does this by packing multiple times at different sizes to find a reasonable solution (packing is NP-hard).
Polygonal packing is much more difficult. Spine does polygonal packing at only one size and even then you probably noticed it takes significantly longer than rectangular packing. It chooses a heuristic and region rotations that produce the smallest packed area within the max size, then it cuts off any extra space, if possible. Because it would take too long to try multiple sizes, in some cases you may achieve a slightly smaller result by using a smaller max size. This isn't always true and even when it is the difference is not typically very much.
Packing your project with 4096x4096 max size: 2233 x 1163 = 2,596,979
Packing your project with 2048x2048 max size: 2048 x 1265 = 2,590,720
Using 2048x2048 packs more efficiently by about 0.24%: 1 - (2,590,720 / 2,596,979) = 0.0024
Note unless you really need power of two textures, you'll get smaller results without it,