I realized afterward that I muddled that. Manifestly, making the baffle wide moves the 1st peak of the diffraction ripple lower in frequency. As such the larger the radius of the round-off relative to the width the baffle the better the effect will be, the ideal being a cylinder with just enough flat region at the face to mount the drivers. For a driver mounted equidistant from both edges the 1st peak will occur at wavelength matching the baffle width. For a driver mount 2/3 of the way from the further edge, that distance will account for 180 degrees (the other 180 degrees is due to the 180 degree phase shift in the "soft" reflection), which means that 2/3 of the baffle width will equal 1/2 wavelength, and that the wavelength will equal 4/3 of the baffle width. For full width equal to 54 cm, if the affected driver is mounted at the midpoint the 1st peak will occur at 635 Hz. If the driver is mounted 2/3 of the way from the further edge (so that the harmonically related peaks occurring higher in frequency than the 1st peak will line up with the dips, i.e., with the frequencies where the wave coming off the edge is out of phase with the direct wave), the 1st peak will occur at 476 Hz. Given the low crossover point you indicated, the affected driver will be the midrange. Since the round-off will just barely be effective at these frequencies, the implication is that the midrange and tweeter should be placed off-center, such that the distance from one edge is twice greater than the distance from the other edge.
I've done this simple analysis a number of times before, and it always comes out this way. This was Linkwitz's point about trying to using rounded edges to overcome the diffraction ripple. In order for it to work, the enclosure pretty much needs to be a cylinder, i.e., the round-off radius has to be not much less than half the baffle width. When this is the case the diffraction ripple is avoided but of course the overall step remains. And to analyze the effect of the step in any meaningful way you have to make assumptions about the speaker placement, especially the distance from the baffle to the wall behind the speaker. The reflection from the wall is sort of the inverse of the baffle step with diffraction ripple. At very low frequency the wall reflection is in phase with the wave coming directly from the driver (which is why it doesn't make sense to try for 6 dB of baffle step correction, contrary to what a lot of hobbyists seem to believe). As the wavelength shortens the reflection from the wall becomes increasingly out of phase with the wave coming directly from the driver, until eventually it is 180 degrees out of phase and a dip occurs in the sound we hear.
An obvious question is whether it is possible to force the frequency where this dip occurs to line up with the first peak in the diffraction ripple. It turns out that the only solution to the equation you end up with is for the baffle to be mounted flush with the wall. This is an interesting exercise for the purpose of demonstrating that this is the only solution. The dip will occur when the total distance travelled by the wave, from the center of the driver to the edge of the baffle, then to the wall, then to the listener, is greater by one-half wavelength than the direct distance from the driver to the listener. For a driver mounted equidistant from both edges, the wavelength for the dip will be equal to the baffle width plus 4x the distance from the baffle edge to the wall. The wavelength of the first peak in the diffraction ripple will be equal to the baffle width, for a driver mounted equidistant from both edges. Thus, the dip will always occur at wavelength greater than the wavelength for the first peak in the diffraction ripple, the difference between the two wavelengths being equal to four times the distance from the baffle edge to the wall. The more closely the speaker is placed to the wall behind it, the more closely the dip will line up with the first peak in the diffraction ripple. If the speaker is mounted flush with the wall there is no baffle step and no diffraction ripple. Of course there is a popular notion that a speaker is best placed out away from the wall, but notwithstanding the popularity of this notion, I have personally not ever encountered an explanation that makes sense. If in fact a speaker sounds better when placed further from the wall, the reason is most likely that the speaker has a poor off-axis response which is rendered less obvious by moving the speaker out away from the wall.