Sir, yes, Sir.
You were right about kernels.
Thanks.

It really looks like multiboot used an older release in its deb. I never ever thought about that But both kernels differ but have about the same creation date, so it was not obvious (to me).
Fetched pr1.3 kernel image, going to create uImage and test (but need to free some space on rootfs aforehand).
Does not yet explain behaviour (or was pr1.2 known to not boot without 0x400000?) and especially why it now worked on my device. But who cares?

Thanks, once more!

btw: I use u-boot to load kernels and multiboot in second stage confirms. If I like to boot another kernel I boot it with u-boot and select it also in multiboot. This way multiboot does not flash but just uses it. Going back to default use the same procedure and kernel does not get flashed! I have fbcon enabled and can select a shell in multiboot. Quite nice and easy to flash kernel on boot (if really needed).