[jbarberu]

Having implemented a custom matrix in qmk, I can say the timescale involved is not something you'd notice a difference between diodes vs diodeless. On ARM boards there's a setup delay to let the gpio propagate before reading the columns, which by default is set to 0.25us, which means reading 4 rows would take about 1us.

I built a macropad that has some regular keys and a SNES controller on the "same matrix", total scan time is 335us, which is dead slow as far as scanning goes, but nothing you'd be able to notice.

Other points are fair though, fewer parts in the BOM. Though routing is a bit trickier with diodeless, especially if you also want to adressable RGB.