diff --git a/src/tcpkali_connection.h b/src/tcpkali_connection.h
index 7ad3b9a..0e0b58b 100644
--- a/src/tcpkali_connection.h
+++ b/src/tcpkali_connection.h
@@ -20,6 +20,7 @@
 struct connection {
     tk_io watcher;
     tk_timer timer;
+    double timer_deadline;
     off_t write_offset;
     struct transport_data_spec data;
     non_atomic_traffic_stats traffic_ongoing;  /* Connection-local numbers */
diff --git a/src/tcpkali_engine.c b/src/tcpkali_engine.c
index 164c447..38a4f88 100644
--- a/src/tcpkali_engine.c
+++ b/src/tcpkali_engine.c
@@ -1474,6 +1474,7 @@ conn_timer_cb(TK_P_ tk_timer *w, int UNUSED revents) {
     struct connection *conn =
         (struct connection *)((char *)w - offsetof(struct connection, timer));
 
+    conn->timer_deadline = 0;
     switch(conn->conn_state) {
     case CSTATE_CONNECTED:
         switch(conn->conn_type) {
@@ -1568,8 +1569,6 @@ static void
 connection_timer_refresh(TK_P_ struct connection *conn, double delay) {
     struct loop_arguments *largs = tk_userdata(TK_A);
 
-    tk_timer_stop(TK_A, &conn->timer);
-
     switch(conn->conn_state) {
     case CSTATE_CONNECTED:
         /* Use the supplied delay */
@@ -1579,7 +1578,21 @@ connection_timer_refresh(TK_P_ struct connection *conn, double delay) {
         break;
     }
 
+    double now = tk_now(TK_A);
+
+    /*
+     * Do nothing if we already have a timer which
+     * is going to be fired sooner than the new one
+     */
+    if(conn->timer_deadline > 0 && conn->timer_deadline < now + delay) {
+        return;
+    }
+
+    tk_timer_stop(TK_A, &conn->timer);
+    conn->timer_deadline = 0;
+
     if(delay > 0.0) {
+        conn->timer_deadline = now + delay;
 #ifdef USE_LIBUV
         uv_timer_init(TK_A_ & conn->timer);
         uint64_t uint_delay = 1000 * delay;
@@ -1602,6 +1615,7 @@ common_connection_init(TK_P_ struct connection *conn, enum conn_type conn_type,
 
     conn->conn_type = conn_type;
     conn->conn_state = conn_state;
+    conn->timer_deadline = 0;
 
     maybe_enable_dump(largs, conn_type, sockfd);
 
@@ -1936,10 +1950,11 @@ static enum lb_return_value {
                  * So if the reading allowance is too small, make it
                  * large enough still to batch reads.
                  */
-                if(allowed_to_move > 1460)
+                if(limit.bytes_per_second < allowed_to_move) {
                     *suggested_move_size = allowed_to_move;
-                else
-                    *suggested_move_size = 1460;
+                } else if(limit.bytes_per_second < *suggested_move_size) {
+                    *suggested_move_size = limit.bytes_per_second;
+                }
                 return LB_PROCEED;
             }
         } else {
@@ -1960,6 +1975,7 @@ static enum lb_return_value {
         }
 
         if(delay < 0.001) delay = 0.001;
+        if(delay > 1) delay = 1;
 
         connection_timer_refresh(TK_A_ conn, delay);
 
@@ -2504,6 +2520,7 @@ connection_cb(TK_P_ tk_io *w, int revents) {
          * If there's nothing to write, we remove the write interest.
          */
         tk_timer_stop(TK_A, &conn->timer);
+        conn->timer_deadline = 0;
         if((conn->data.total_size == 0) && !(conn->conn_blocked & CBLOCKED_ON_WRITE)) {
             conn->conn_wish &= ~CW_WRITE_INTEREST; /* Remove write interest */
             update_io_interest(TK_A_ conn);