Ok, after some fooling around this weekend I think I have a solution that will be satisfactory for me.
I’ve published all my files and OpenSCAD source code on github:
I’ve ended up going with the magnet based solution and it seems to work really well. Using 4 pot magnets the beds attach with enough force so that I can lift the machine up by picking it up by the bedplate. I assume that should be strong enough 
To prevent it from sliding sideways, I’ve embedded the magnet and metal counter discs into a “pot” and “lid” which snap together perfectly and consistently in the same space as well.
So far I’m really pleased with this solution. My proof of concept version works great. Now I just have to wait for my order of inserts to arrive so I can create the final versions of the top-boards.
A video of it in action (preview should work now):
https://1drv.ms/v/s!ArWNERowM_MXrVhKSm1dD6fUEsuZ?e=Vewce6
(yes it’s the experimental version and a leftover piece of wood used. Final version will be a full plate obviously :))
Some pictures, also included on the github page:
biggest disadvantage: losing up to 36mm of height. This can be rather easily reduced to max 26mm by using 10mm MDF in stead of 18mm MDF for the top plates. It could probably be reduced even more by countersinking the carriage in the bottom plate. (mounting directly on the Y-axes isn’t a viable option I think)
But for me personally, it’s a sacrifice I’m willing to make. And if needed, it’s still removable. Up till now I didn’t encounter any project yet where I needed to go that high. (Not yet at least, I’m sure this will come back to me as a boomerang really fast 
)