You could trade off some security for speed like that, but 100 rows wouldn’t cut it. Ask for the same information again and you’ll probably only have 1 intersection. Even if you read half of the rows unnecessarily, you could figure out the true row in between 2 and approximately log2(rows) reads. Just isn’t worth the side channel attack.
No, if you run the same query the server must not be able to determine that it was the same query. Repeated invocations of the same question would be transformed with added noise during encryption so that the encrypted query and answer is different each time.
If the same query would access the same 100 random rows, then after seeing your (hidden) query and getting the 100-random-row signature it would be trivial to run a bunch of candidates and see which one of them gets the same 100-random-row signature and thus figure out what you queried.
