I wonder whether long polling doesn't work for me. I've got a rather high stale ratio (about 4-5%) and there're always a few rejected shares reported by phoenix just about the time there's a new block solved on DeepBit. E.g. (times are GMT+4)

[24/05/2011 14:18:41] Result: bd7a10c0 accepted
[24/05/2011 14:19:02] Result: fc5adf2a rejected
[24/05/2011 14:19:04] Result: 477f3c89 rejected
[24/05/2011 14:19:18] Result: 2e53a343 accepted
...
[24/05/2011 14:26:02] Result: f06cc602 accepted
[24/05/2011 14:26:09] Result: ef89fce7 rejected
[24/05/2011 14:26:09] Result: c055e173 rejected
[24/05/2011 14:26:15] Result: c2d84f2e rejected
[24/05/2011 14:26:17] Result: 30afeee2 rejected
[24/05/2011 14:27:00] Result: 692ce319 accepted
...
[24/05/2011 14:38:53] Result: d132f55c accepted
[24/05/2011 14:39:35] Result: 3c1b5992 rejected
[24/05/2011 14:39:36] Result: 0163f921 rejected
[24/05/2011 14:39:42] Result: f76ef18c rejected
[24/05/2011 14:40:01] Result: 6e0fa1b0 accepted
...
[24/05/2011 15:29:51] Result: e2671d16 accepted
[24/05/2011 15:30:06] Result: 2d1ca984 rejected
[24/05/2011 15:30:19] Result: 3487f2e0 accepted

Looks to me as if the miner doesn't receive notifications from the pool and continues to calculate an outdated block. I'm using phoenix 1.48 with -u user@mail.com:1234567890@deepbit.net:8332/ -k phatk DEVICE=0 VECTORS BFI_INT WORKSIZE=128 AGGRESSION=8
Hash rate is about 228Mh/s on HD6850 @ 875