The author dealt with this outside the article, and posted a link to his slides in this HN post. The relevant slides begin at [1].
At the end of the day a straight cut is limiting. The next step would be to design the perfect onion dicing knife.
Those slides show that the solution won't work on actual onions. Call the innermost layer of the onion its "biological center", and call the center of the spheroid approximately occupied by the onion its "geometrical center".
As is beautifully illustrated on slide 50, the biological center is generally not particularly close to the geometrical center, and this introduces huge distortions in slices that cut close to the biological center.* A single layer of the onion can run parallel to the knife cut for quite some distance.
* The slides also observe that in reality, before chopping an onion, you cut off the top and bottom. This same phenomenon explains why you have to do that; a vertical cut through the top or bottom end of the onion would just give you one huge piece. (You also need to get rid of the roots on the bottom and the sprouts on the top, but even if you didn't, you'd have to cut off the top and the bottom because they curve the wrong way.)
As you point out, without a perfectly symmetrical onion of course this will not work very well. You would need to use a moving geometrical center point when slicing for best results.
Further, as noted elsewhere the outer layers are thicker in a real onion, so we need to reformulate to take this into account.
The other obvious simple improvement I can think of would be to use radial cuts in both directions. Each direction with the its own optimized floating center point of course. Reformulations would need to take this into account - although the end result would be quite close, and likely well within the margin of error for almost any human being aiming at an imagined floating center point below a cutting board :).