AFAIK Lightroom does not rely on filenames to detect potential duplicates. Files are hashed and any matching file signature will be declared as duplicate. This is the reason why the import takes such a long time to find matches: all files on the source / import card need to be read and hashed to find duplicates.
One use case where files would not match is if you are transforming RAW files into DNG files, which then get changed after the import (new metadata added to the DNG files). This assumes that the whole file is hashed and not just the image part, I don't know if this is the way Lr works.
I would try to manually produce hashes of file pairs to assess if the files are really identical. Any programme producing md5, sha1 or sha256 hashes could be used for that purpose. That would at least rule this potential issue out.
Cheers,
Fabien