diff --git a/jemalloc/doc/jemalloc.3.in b/jemalloc/doc/jemalloc.3.in
index efe773bd..f55325e8 100644
--- a/jemalloc/doc/jemalloc.3.in
+++ b/jemalloc/doc/jemalloc.3.in
@@ -312,6 +312,15 @@ Double/halve the size of the maximum size class that is a multiple of the
 quantum (8 or 16 bytes, depending on architecture).
 Above this size, cacheline spacing is used for size classes.
 The default value is 128 bytes.
+@roff_tcache@.It S
+@roff_tcache@Sort the objects of a particular size class that are stored in a
+@roff_tcache@thread-specific cache just before flushing some of them from the
+@roff_tcache@cache, such that the objects highest in memory are preferentially
+@roff_tcache@freed.
+@roff_tcache@This tends to reduce fragmentation, but sorting is (n lg n), and in
+@roff_tcache@practice it is expensive enough to have a moderate performance
+@roff_tcache@impact.
+@roff_tcache@This option is enabled by default.
 @roff_trace@.It T
 @roff_trace@Write a verbose trace log to a set of files named according to the
 @roff_trace@pattern
diff --git a/jemalloc/src/jemalloc.c b/jemalloc/src/jemalloc.c
index 71db5199..9fe90423 100644
--- a/jemalloc/src/jemalloc.c
+++ b/jemalloc/src/jemalloc.c
@@ -1090,6 +1090,7 @@ static bool	opt_junk = false;
 #ifdef JEMALLOC_TCACHE
 static size_t	opt_lg_tcache_nslots = LG_TCACHE_NSLOTS_DEFAULT;
 static ssize_t	opt_lg_tcache_gc_sweep = LG_TCACHE_GC_SWEEP_DEFAULT;
+static bool	opt_tcache_sort = true;
 #endif
 static ssize_t	opt_lg_dirty_mult = LG_DIRTY_MULT_DEFAULT;
 static bool	opt_stats_print = false;
@@ -4058,7 +4059,7 @@ tcache_bin_flush(tcache_bin_t *tbin, size_t binind, unsigned rem)
 	void *ptr;
 	unsigned i, ndeferred, ncached;
 
-	if (rem > 0) {
+	if (opt_tcache_sort && rem > 0) {
 		assert(rem < tbin->ncached);
 		/* Sort pointers such that the highest objects will be freed. */
 		tcache_bin_sort(tbin);
@@ -5696,6 +5697,14 @@ MALLOC_OUT:
 					    opt_lg_cspace_max)
 						opt_lg_qspace_max++;
 					break;
+#ifdef JEMALLOC_TCACHE
+				case 's':
+					opt_tcache_sort = false;
+					break;
+				case 'S':
+					opt_tcache_sort = true;
+					break;
+#endif
 #ifdef JEMALLOC_TRACE
 				case 't':
 					opt_trace = false;
@@ -6418,6 +6427,9 @@ malloc_stats_print(const char *opts)
 		malloc_message(opt_junk ? "J" : "j", "", "", "");
 #endif
 		malloc_message("P", "", "", "");
+#ifdef JEMALLOC_TCACHE
+		malloc_message(opt_tcache_sort ? "S" : "s", "", "", "");
+#endif
 #ifdef JEMALLOC_TRACE
 		malloc_message(opt_trace ? "T" : "t", "", "", "");
 #endif