diff --git a/src/test/bench/msgsend.rs b/src/test/bench/msgsend.rs
new file mode 100644
index 00000000000..bae12896a78
--- /dev/null
+++ b/src/test/bench/msgsend.rs
@@ -0,0 +1,60 @@
+// A port of the simplistic benchmark from
+//
+//    http://github.com/PaulKeeble/ScalaVErlangAgents
+//
+// I *think* it's the same, more or less.
+
+use std;
+import std::io::writer;
+import std::io::writer_util;
+
+enum request {
+    get_count,
+    bytes(uint),
+    stop
+}
+
+fn server(requests: comm::port<request>, responses: comm::chan<uint>) {
+    let count = 0u;
+    let done = false;
+    while !done {
+        alt comm::recv(requests) {
+          get_count { comm::send(responses, copy count); }
+          bytes(b) { count += b; }
+          stop { done = true; }
+        }
+    }
+    comm::send(responses, count);
+}
+
+fn run(args: [str]) {
+    let server = task::spawn_connected(server);
+    let size = uint::from_str(args[1]);
+    let workers = uint::from_str(args[2]);
+    let start = std::time::precise_time_s();
+    let to_child = server.to_child;
+    let worker_tasks = [];
+    uint::range(0u, workers) {|_i|
+        worker_tasks += [task::spawn_joinable {||
+            uint::range(0u, size / workers) {|_i|
+                comm::send(to_child, bytes(100u));
+            }
+        }];
+    }
+    vec::iter(worker_tasks) {|t| task::join(t); }
+    comm::send(server.to_child, stop);
+    let result = comm::recv(server.from_child);
+    let end = std::time::precise_time_s();
+    let elapsed = end - start;
+    std::io::stdout().write_str(#fmt("Count is %?\n", result));
+    std::io::stdout().write_str(#fmt("Test took %? seconds\n", elapsed));
+    let thruput = (size / workers * workers as float) / (elapsed as float);
+    std::io::stdout().write_str(#fmt("Throughput=%f per sec\n", thruput));
+}
+
+fn main(args: [str]) {
+    let args1 = if vec::len(args) <= 1u { ["", "10000", "4"] } else { args };
+    #debug("%?", args1);
+    run(args1);
+}
+