This is why Pi = 2 is not a good idea
The easiest solution is to house-rule diagonals back as they are in 3.x. Otherwise, you're stuck in a geometry that has a Pi of 2, where circles are squares, and any accurate visualization of even this trivial of a problem requires double the number of dimensions one would expect (4 in this case).
Using a crude method of estimation, I think that a diagonal "square" that's length roughly 2.12 (=3/sqrt(2)) on the right side (why? Because in that direction, sqrt(2) = 1 in the other direction. really - you need to look at this problem in 4 dimensions. Make a bunch of circles in paper, and tape them together so that each circle touches 8 others, and has no empty space between circles, and you'll quickly see what I mean as you'll run out of 3-space very quickly) would get you close to the same area as the one on the left side.
Of course, this is sort of a meaningless exercise - why? because the square you show on the left side of your example is, in fact, diagonal. Even worse, if you rotate it any number of degrees, you get exactly the same shape as the one on the left side; the one on the right is no form of rotation, of the one on the left. That's because a square is a circle. From a geometric standpoint, what you're trying to do on the right side is the same as offsetting a area effect in 3.x so that it hits different partial squares. As such, only portions of some squares are hit.
So if I were to allow offsets, I'd permit a diagonal of 2 (why? because 2 is a simpler number than 2.12), which would likely mean that 2 orcs is are fully hit, and 2 orcs are hit in half their space (perhaps give them +2 to defense and evasion?).
Now, obviously I'm ignoring RAW above. If you do permit the shape on the right, I'd consider virtually always using it - it simply hits more area. If you want an argument to disallow it, note that the shape on the right is not a square.