The privacy gain is a nice side effect of using probabilistic filters but not main benefit as far as I'm concerned. the benefit to using them is allowing clients to quickly check if a server says its hosting blobs.
The existing HEAD /<sha256> endpoint is a nice way to check the status of a single hash. but when the client needs to check >100 hashes it get really spamy. probabilistic filters could allow clients to check the status of almost an unlimited number of hashes with a single request.