We should use "large" instance types for the concurrency_replication_causal_consistency suites. They start 5 nodes that each have a 1GB oplog and 1GB WT cache, this overwhelms our "small" instances which have 7.5GB memory. We workaround the problem at the moment by running CleanEveryN to remove the data files and restart the cluster, so the oplog doesn't get too big.

Similarly, concurrency sharded and replication suites on the linux repeated builder should also run on the large instance type.