We're observing a number of BFs on aggregation_mongos_passthrough task due, probably to the fact that https://github.com/10gen/mongo/commit/88e5f1ba748e7df3611507eeb9dcbfbd4dc7f6fb introduces 2 (to a total of 3) additional mongos to each sharded deployment.

This is causing issues where, apparently, one mongos still considers the collection unsharded and hence, pipelines are not being optimized in cluster_aggregate.cpp. optimizePipeline()

Currently worked around with assert.soon to wait for routing table availability but that could have customer impact of running unoptimized queries.