I'm also seeing something where the size of the queue on each GPU keeps growing over time, until there are hundreds of queued work per card. It may have something to do with cgminer reporting that the pool isn't providing work quickly enough (which I think is incorrect, the pushpoold it's talking to is a few feet away, on a totally unloaded box, that older versions of cgminer aren't complaining about), and after that the size of the queue keeps growing like crazy.
This also seems to be causing some of the work that's being done to be so old it's being rejected by the time it actually gets a chance to run. Example from a box running for a few hours:
[(5s):845.3 (avg):833.2 Mh/s] [Q:1336 A:1820 R:84 HW:0 E:136% U:10.99/m]
TQ: 8 ST: 8 SS: 0 DW: 11 NB: 18 LW: 2745 GF: 4 RF: 0 I: 9
Connected to
Block: 00000938f98c268dcf86a8cb4efa000a... Started: [18:28:21]
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0: [95.0 C] [367.3/373.5 Mh/s] [Q:673 A:810 R:43 HW:0 E:120% U:4.89/m]
GPU 1: [88.0 C] [187.7/186.9 Mh/s] [Q:251 A:400 R:11 HW:0 E:159% U:2.41/m]
GPU 2: [99.0 C] [106.1/100.8 Mh/s] [Q:137 A:235 R:9 HW:0 E:172% U:1.42/m]
GPU 3: [88.0 C] [179.8/172.1 Mh/s] [Q:225 A:375 R:21 HW:0 E:167% U:2.26/m]
That's total queued to date, not current queued.