By inspection of its source code, abootimg has the same bug as Android mkbootimg: because it writes host structs directly into the file, if the size or endianness of any of the integers in those structs does not match the expected size/endianness (4-byte int, little-endian), the image will be misinterpreted by the Android device. Another possibility would be to only build abootimg on little-endian platforms (any-i386, any-amd64, armel, armhf, mipsel etc., but not powerpc).
I agree, but as I don't have any big-endian system under my hand, I won't be able to test it. Patches are welcome ;) It would make sense. I don't see any usage case where you would like to manipulate Android boot image from a bigendian host... Thanks, Gilles.