I have done this before and it worked quite well.
AWS tends to have very good latency between regions and a hugely redundant network so replication lag is not a problem but you might wish to extend the capped collection oplog a little bit to account for regional problems (i.e. Hurricane Sandy in one region and sunny weather in the other).
As @Orid said you will, of course, want to ensure CAP by keeping a majority in at least one side of the network partition which will exist between regions.