It's pretty similar, but not exactly like that. The main difference is that the server doesn't need to transmit a bloom filter, it transmits partial hashes, but yes, there's a tradeoff.
> How many 'longer hashes' do clients get on average per submitted 'short hash'? How large is the 'false positive' rate you get from the long hashes?
I'm afraid I don't know the answer to these questions, as we don't keep logs... The false positive rate of the long hashes should be astronomically low (since they are full 256-bit hashes), but it doesn't much matter because the client only needs to be reasonably sure that the user exists so it can perform an actual query with the plaintext identifier.
The problem isn't really "we shouldn't know any of your contacts", it's "we shouldn't know any of your contacts that aren't using our service", since clients have to divulge some of their contacts in order to set up calls with them or send them messages. That means the client only has to know which of their contacts are probably on the server so they can request more information, which will, at that point, tell them with certainty whether a contact is using the service.
P.S. My opinions are my own, I'm not speaking for SC, etc.