When I use cassandra-cli to execute

get InfinispanExpiration[utf8('expiration')];

I get all 3099 entries in 1990ms.

All super columns contain only one column (so all entries have a different expiration time)

=> (super_column=1340711171661,
(column=656e7472795f687474703a2f2f31302e3233332e32332e3132343a383038302f68697434636c6c2f..., value=, timestamp=1340624772033000))
=> (super_column=1340711172132,
(column=656e7472795f687474703a2f2f31302e3233332e32332e3132343a383038302f68697434636c6c2f..., value=, timestamp=1340624772298000))
=> (super_column=1340711207243,
(column=656e7472795f687474703a2f2f31302e3233332e32332e3132343a383038302f68697434636c6c2f..., value=, timestamp=1340624808035000))
=> (super_column=1340711262809,
(column=656e7472795f687474703a2f2f31302e3233332e32332e3132343a383038302f68697434636c6c2f..., value=, timestamp=1340624864878000))
...

(The entry keys are encoded uris, about 200 to 300 bytes long)

So I don't think my data is the problem.
I had a quick look at the code of CassandraCacheStore and I didn't see a problem there either. Maybe it is the underlying thrift library?