If the plot were smaller there's the risk that someone could calculate plots "on the fly" within the 30-second claim window, thus reducing Chia to proof-of-work (or even breaking it entirely? the "green" paper [1] is a bit too dense for me). It's expected and planned that k will have to increase to 33 in a few years to stay ahead.
[1]: https://www.chia.net/assets/ChiaGreenPaper.pdf