diff --git a/.github/workflows/c-cpp-Linux.yml b/.github/workflows/c-cpp-Linux.yml
index c906317..670d57e 100644
--- a/.github/workflows/c-cpp-Linux.yml
+++ b/.github/workflows/c-cpp-Linux.yml
@@ -30,6 +30,8 @@ jobs:
       run: make -f Makefile.Linux MAILPROXY=true FTP=true
     - name: regression tests
       run: python3 tests/run.py
+    - name: regression tests, a thread for each client
+      run: python3 tests/run.py --loopmode off
     - name: mkdir
       run: mkdir ~/3proxy
     - name: make install
diff --git a/man/3proxy.cfg.5 b/man/3proxy.cfg.5
index d1ffb0d..98572d1 100644
--- a/man/3proxy.cfg.5
+++ b/man/3proxy.cfg.5
@@ -254,9 +254,10 @@ authorization and to drop the per-destination log records only.
  (for udppm) singlepacket, expect only one packet from both client and server
 .br
 .B -L
- (for TCP services) serve the clients on event loops (see \fBloopmode\fR): a
-few threads each serve many clients, instead of a thread for each.
-\fB-L0\fR gives each client a thread of its own whatever \fBloopmode\fR says.
+ (for TCP services) how the clients are served, whatever \fBloopmode\fR says
+(see there): \fB-L2\fR on event loops, which also relay their data (the
+default), \fB-L\fR on event loops relaying the data as a thread would,
+\fB-L0\fR a thread for each client.
 .br
 .B -u
 Never ask for username/password
@@ -1526,16 +1527,21 @@ many connections were given to a waiting one on its Running Services page.
 
 .br
 .BR loopmode
-\fIon\fR|\fIoff\fR
+\fIcb\fR|\fIon\fR|\fIoff\fR
 .br
  Whether the clients of the TCP services which follow (until the next
 \fBloopmode\fR) are served on event loops: a few threads (see \fBloops\fR)
 each wait for the sockets and timers of many clients and switch between
 them, each client running on a stack of its own as large as its thread's
 would be. A client costs a stack and no thread, and waking it costs no
-switch of threads, so many connections take less memory and CPU. Off by
-default; the \fB-L\fR option of a service overrides it, \fB-L0\fR for a
-thread for each client.
+switch of threads, so many connections take less memory and CPU.
+\fIcb\fR, the default, also has the loop relay a connection's data itself
+once the request is handled, where nothing but the sockets is between the
+sides (no filters, no limits but that of a body's length), without
+switching to the client's stack for each read. \fIon\fR relays the data
+on the client's stack, as a thread of its own would, and \fIoff\fR gives
+each client a thread. The \fB-L\fR option of a service overrides it: \fB-L2\fR,
+\fB-L\fR, \fB-L0\fR.
 .br
  Clients keep a thread each, whatever is asked, where something they use
 would hold up the others sharing their thread: a service with FilePlugin,
diff --git a/src/conf.c b/src/conf.c
index d7086a8..ecb3854 100644
--- a/src/conf.c
+++ b/src/conf.c
@@ -429,7 +429,7 @@ static int h_loops(int argc, unsigned char **argv){
 static int h_loopmode(int argc, unsigned char **argv){
 	if(!strcmp((char *)argv[1], "on")) conf->loopmode = 1;
 	else if(!strcmp((char *)argv[1], "off")) conf->loopmode = 0;
-	/* and their data relayed by the loop (a draft) */
+	/* and their data relayed by the loop: the default */
 	else if(!strcmp((char *)argv[1], "cb")) conf->loopmode = 2;
 	else return 1;
 	return 0;
diff --git a/src/confinit.h b/src/confinit.h
index de1c91e..af05924 100644
--- a/src/confinit.h
+++ b/src/confinit.h
@@ -22,6 +22,8 @@
 	.trafcounter = NULL,
 	.services = NULL,
 	.stacksize = 0,
+	/* clients served on event loops, their data relayed by the loops */
+	.loopmode = 2,
 	.poolmax = DEFAULT_POOLMAX,
 	.pooltimeout = DEFAULT_POOLTIMEOUT,
 	.counterd = -1,
diff --git a/src/mainconf.c b/src/mainconf.c
index 0dad946..c2d3732 100644
--- a/src/mainconf.c
+++ b/src/mainconf.c
@@ -480,7 +480,7 @@ static int srvprepare(struct srvstart *s, int argc, char **argv){
 			srv->stacksize = atoi(argv[i]+2);
 			break;
 		 case 'L':
-			/* -L2: and their data relayed by the loop (a draft) */
+			/* -L2: and their data relayed by the loop (the default) */
 			srv->loopmode = argv[i][2]? atoi(argv[i]+2) : 1;
 			break;
 		case 'a':
diff --git a/src/mainloop.c b/src/mainloop.c
index a539e8e..dcc76d3 100644
--- a/src/mainloop.c
+++ b/src/mainloop.c
@@ -7,9 +7,10 @@
 */
 
 /* Clients served on event loops: a few threads, a CPU each by default
-   (loops), each running a loop, and each client of a service given -L (or
-   loopmode on) served by a fiber of one of them, with the same code as a
-   client with a thread of its own: its waits in sockpoll() wait in the loop.
+   (loops), each running a loop, and each client of a service (unless given
+   -L0 or loopmode off) served by a fiber of one of them, with the same code
+   as a client with a thread of its own: its waits in sockpoll() wait in the
+   loop.
    A service whose clients cannot share a thread (threadperclient), one
    listening for connections back (-R), all of its clients waiting for the
    same socket, which a loop waits for once, or a build or processor without
@@ -426,13 +427,13 @@ DONE:
 }
 
 
-/* ---- sockmap() as callbacks (-L2, a draft) ------------------------------- */
+/* ---- sockmap() as callbacks (-L2, loopmode cb) --------------------------- */
 
 /* The data of a connection mapped by a fiber of a loop with neither filters,
    limits but that of a body's length, nor anything but the socket functions of the system between the
    sides, is relayed by the loop itself: the sockets call it back, on the
    loop's own stack, and the fiber, parked meanwhile, is resumed only once
-   the mapping is over. No splice() yet: data is copied. */
+   the mapping is over. */
 
 /* rounds of a read and a write each way at most in a callback, before what
    is still ready waits for the next turn of the loop */
diff --git a/src/structures.h b/src/structures.h
index df20696..b146e0a 100644
--- a/src/structures.h
+++ b/src/structures.h
@@ -1009,7 +1009,8 @@ struct extparam {
 	int filtermaxsize;
 	int poolmax, pooltimeout;	/* threadpool: idle workers kept, -1 automatic */
 	int loops;		/* threads serving clients on event loops, 0: a CPU each */
-	int loopmode;		/* services from here on serve clients on loops */
+	int loopmode;		/* services from here on serve clients on loops: 1, and
+				   relay their data there: 2 */
 	int gracetraf, gracenum, gracedelay;
 	int maxseg;
 	unsigned char **archiver;
diff --git a/tests/run.py b/tests/run.py
index 6e7db07..152e762 100644
--- a/tests/run.py
+++ b/tests/run.py
@@ -58,11 +58,14 @@ def main():
                         help="keep the temporary directory")
     parser.add_argument("-v", "--verbose", action="store_true",
                         help="print every check, not just the failures")
-    parser.add_argument("--loopmode", action="store_true",
-                        help="serve the clients of every service on event loops")
+    parser.add_argument("--loopmode", choices=("cb", "on", "off"),
+                        help="serve the clients of every service so (cb, the "
+                        "default, and on: on two event loops; off: a thread "
+                        "each)")
     args = parser.parse_args()
     if args.loopmode:
-        os.environ["THREEPROXY_TESTCONFIG"] = "loops 2\nloopmode on\n"
+        os.environ["THREEPROXY_TESTCONFIG"] = \
+            "loops 2\nloopmode %s\n" % args.loopmode
 
     binary = args.binary or os.environ.get("BIN") or default_binary()
     binary = os.path.abspath(binary)