diff --git a/bench-tps/src/bench.rs b/bench-tps/src/bench.rs
index 6dc44ed1f..0842fa02c 100644
--- a/bench-tps/src/bench.rs
+++ b/bench-tps/src/bench.rs
@@ -233,9 +233,13 @@ pub fn do_tx_transfers(
     leader: &NodeInfo,
     shared_tx_thread_count: &Arc<AtomicIsize>,
     total_tx_sent_count: &Arc<AtomicUsize>,
+    thread_batch_sleep_ms: usize,
 ) {
     let client = mk_client(&leader);
     loop {
+        if thread_batch_sleep_ms > 0 {
+            sleep(Duration::from_millis(thread_batch_sleep_ms as u64));
+        }
         let txs;
         {
             let mut shared_txs_wl = shared_txs.write().unwrap();
diff --git a/bench-tps/src/cli.rs b/bench-tps/src/cli.rs
index 23c87d624..e9e1fb73d 100644
--- a/bench-tps/src/cli.rs
+++ b/bench-tps/src/cli.rs
@@ -15,6 +15,7 @@ pub struct Config {
     pub num_nodes: usize,
     pub duration: Duration,
     pub tx_count: usize,
+    pub thread_batch_sleep_ms: usize,
     pub sustained: bool,
     pub reject_extra_nodes: bool,
     pub converge_only: bool,
@@ -30,6 +31,7 @@ impl Default for Config {
             num_nodes: 1,
             duration: Duration::new(std::u64::MAX, 0),
             tx_count: 500_000,
+            thread_batch_sleep_ms: 0,
             sustained: false,
             reject_extra_nodes: false,
             converge_only: false,
@@ -110,6 +112,14 @@ pub fn build_args<'a, 'b>() -> App<'a, 'b> {
                 .takes_value(true)
                 .help("Number of transactions to send per batch")
         )
+        .arg(
+            Arg::with_name("thread-batch-sleep-ms")
+                .short("z")
+                .long("thread-batch-sleep-ms")
+                .value_name("NUM")
+                .takes_value(true)
+                .help("Per-thread-per-iteration sleep in ms"),
+        )
 }
 
 /// Parses a clap `ArgMatches` structure into a `Config`
@@ -158,6 +168,13 @@ pub fn extract_args<'a>(matches: &ArgMatches<'a>) -> Config {
         args.tx_count = s.to_string().parse().expect("can't parse tx_account");
     }
 
+    if let Some(t) = matches.value_of("thread-batch-sleep-ms") {
+        args.thread_batch_sleep_ms = t
+            .to_string()
+            .parse()
+            .expect("can't parse thread-batch-sleep-ms");
+    }
+
     args.sustained = matches.is_present("sustained");
     args.converge_only = matches.is_present("converge-only");
     args.reject_extra_nodes = matches.is_present("reject-extra-nodes");
diff --git a/bench-tps/src/main.rs b/bench-tps/src/main.rs
index eb3be6de0..e6fef1298 100644
--- a/bench-tps/src/main.rs
+++ b/bench-tps/src/main.rs
@@ -82,6 +82,7 @@ fn main() {
         drone_addr,
         id,
         threads,
+        thread_batch_sleep_ms,
         num_nodes,
         duration,
         tx_count,
@@ -207,6 +208,7 @@ fn main() {
                         &leader,
                         &shared_tx_active_thread_count,
                         &total_tx_sent_count,
+                        thread_batch_sleep_ms,
                     );
                 })
                 .unwrap()