It seems that bumping the versions of tokio-rustls and rustls inadvertently led to a significant reduction in performance for large reads.

View the following patch and highlight the "Read Bench" trendline to see it: https://evergreen.mongodb.com/task/mongo_rust_driver_benchmarking_driver_benchmarks__async_runtime~tokio_auth_and_tls~auth_and_tls_os~ubuntu_18.04_benchmark_5.0_standalone_90bc23ff88cebd015c7ed99c9d551c748ee9dc29_22_03_31_18_51_38##hidden=DriverBench%252CWriteBench%252CReadBench%252CParallelBench%252CMultiBench%252CSingleBench%252CLDJSON%252Bmulti-file%252Bexport%252CLDJSON%252Bmulti-file%252Bimport%252CLarge%252Bdoc%252Bbulk%252Binsert%252CSmall%252Bdoc%252Bbulk%252Binsert%252CFind%252Bmany%252Band%252Bempty%252Bcursor%252CLarge%252Bdoc%252BinsertOne%252CSmall%252Bdoc%252BinsertOne%252CFind%252Bone%252CRun%252BCommand%252CFind%252Bmany%252Band%252Bempty%252Bcursor%252B%2528serde%252Bstructs%2529%252CFind%252Bmany%252Band%252Bempty%252Bcursor%252B%2528raw%252BBSON%2529%252CLDJSON+multi-file+export%252CLDJSON+multi-file+import%252CLarge+doc+bulk+insert%252CSmall+doc+bulk+insert%252CFind+many+and+empty+cursor%252CLarge+doc+insertOne%252CSmall+doc+insertOne%252CFind+one%252CRun+Command%252CFind+many+and+empty+cursor+%2528serde+structs%2529%252CFind+many+and+empty+cursor+%2528raw+BSON%2529&threads=maxonly&selected.DriverBench=MAX&selected.WriteBench=MAX&selected.ReadBench=MAX&selected.ParallelBench=MAX&selected.MultiBench=MAX&selected.SingleBench=MAX&selected.Find+many+and+empty+cursor+%2528serde+structs%2529=MAX&selected.Find+many+and+empty+cursor+%2528raw+BSON%2529=MAX&selected.LDJSON+multi-file+export=MAX&selected.LDJSON+multi-file+import=MAX&selected.Large+doc+bulk+insert=MAX&selected.Small+doc+bulk+insert=MAX&selected.Find+many+and+empty+cursor=MAX&selected.Large+doc+insertOne=MAX&selected.Small+doc+insertOne=MAX&selected.Find+one=MAX&selected.Run+Command=MAX