[XDCR on UPR, internal rep= TAP] item count mismatch: expired items not deleted, shows up in queries, get() returns value

Build
-------
3.0.0-666

Scenario
-------------

1. Bi-XDCR between 3 buckets
2. Load 20K items on both sides. Pause XDCR on both sides during loading.
3. Resume XDCR. Expect 40K on all buckets on both sides after bi-xdcr.
4. Update 30% of cluster keys(on each cluster - non-overlapping keys) with expiration time of 20s
5. Delete 30% of cluster keys(non-overlapping keys)
6. Run expiry pager after 20s. Verify item count and do other validations.

Observation
--------------------
Source:
default - 16005
sasl_bucket_1 16005
standard_bucket_1 - 16008

Dest:
default - 16005
sasl_bucket_1 16002
standard_bucket_1 - 16000

Expected: 16000 in each bucket. In re-runs, I saw item counts less than 16000.
Note: bucket warmup makes no difference in this case.

Also, vb_active_items ! = vb_replica_items
2014-05-12 18:34:41 | WARNING | MainProcess | Cluster_Thread | [task.check] Not Ready: vb_active_curr_items 16001 == 16000 expected on '10.3.4.186:8091''10.3.4.187:8091', standard_bucket_1 bucket
2014-05-12 18:34:41 | WARNING | MainProcess | Cluster_Thread | [task.check] Not Ready: vb_replica_curr_items 16004 == 16000 expected on '10.3.4.186:8091''10.3.4.187:8091', standard_bucket_1 bucket
2014-05-12 18:34:42 | WARNING | MainProcess | Cluster_Thread | [task.check] Not Ready: curr_items 16005 == 16000 expected on '10.3.4.186:8091''10.3.4.187:8091', default bucket

Source : http://172.23.106.45:8091/
Destination : http://172.23.106.47:8091/

Testcase
--------------

./testrunner -i bixdcr.ini -t xdcr.pauseResumeXDCR.PauseResumeTest.replication_with_pause_and_resume,items=20000,rdirection=bidirection,ctopology=chain,standard_buckets=1,expires=20,pause=source-destination,doc-ops=update-delete,doc-ops-dest=update-delete

Reproducible
--------------------
Yes, consistently, 4 on 4 times.