benchRun results might be slightly incorrect as long as the tool relies on serverStatus() calls rather than collecting stats client side. For the same reason (stats collected server side) two parallel executions of benchRun won't return per-client results (but cumulative results).

As an added bonus, scotthernandez noticed that if the test lasts < 1 second, the results returned are completely broken, as showed in this testcase (provided by him).

$ mongo --eval 'printjson(benchRun( { ops : [{op:"command", ns:"test",
command:{ping:1}}] , parallel : 1 , seconds : .001 , host : db.getMongo().host }
))'
MongoDB shell version: 2.6.1
connecting to: test
{
...	"command" : 9000
}