diff --git a/libhdf5/hdf5var.c b/libhdf5/hdf5var.c
index 789ad6e34..212521508 100644
--- a/libhdf5/hdf5var.c
+++ b/libhdf5/hdf5var.c
@@ -777,17 +777,23 @@ int
 NC4_def_var_deflate(int ncid, int varid, int shuffle, int deflate,
                     int deflate_level)
 {
-    int stat = NC_NOERR;
+    int stat;
     unsigned int level = (unsigned int)deflate_level;
-    /* Set shuffle first */
-    if((stat = nc_def_var_extra(ncid, varid, &shuffle, NULL, NULL, NULL,
-				NULL, NULL, NULL, NULL, NULL, NULL, NULL))) goto done;
-    if(deflate) {
-        if((stat = nc_def_var_filter(ncid, varid, H5Z_FILTER_DEFLATE,1,&level))) goto done;
-    } /* else ignore */
 
-done:
-    return stat;
+    /* Set shuffle first */
+    if ((stat = nc_def_var_extra(ncid, varid, &shuffle, NULL, NULL, NULL,
+                                 NULL, NULL, NULL, NULL, NULL, NULL, NULL)))
+        return stat;
+
+    /* Don't turn on deflate if deflate_level = 0. It's a valid zlib
+     * setting, but results in a write slowdown, and a file that is
+     * larger than the uncompressed file would be. So when
+     * deflate_level is 0, don't use compression. */
+    if (deflate && deflate_level)
+        if ((stat = nc_def_var_filter(ncid, varid, H5Z_FILTER_DEFLATE, 1, &level)))
+            return stat;
+
+    return NC_NOERR;
 }
 
 /**
diff --git a/nc_perf/Makefile.am b/nc_perf/Makefile.am
index 90f588279..8e29197be 100644
--- a/nc_perf/Makefile.am
+++ b/nc_perf/Makefile.am
@@ -26,7 +26,8 @@ LDADD = ${top_builddir}/liblib/libnetcdf.la
 check_PROGRAMS = tst_create_files bm_file tst_chunks3 tst_ar4	\
 tst_ar4_3d tst_ar4_4d bm_many_objs tst_h_many_atts bm_many_atts	\
 tst_files2 tst_files3 tst_mem tst_mem1 tst_knmi bm_netcdf4_recs	\
-tst_wrf_reads tst_attsperf bigmeta openbigmeta tst_bm_rando	
+tst_wrf_reads tst_attsperf bigmeta openbigmeta tst_bm_rando	\
+tst_compress
 
 bm_file_SOURCES = bm_file.c tst_utils.c
 bm_file_LDFLAGS = -no-install
@@ -41,6 +42,7 @@ tst_h_many_atts_SOURCES = tst_h_many_atts.c tst_utils.c
 tst_knmi_SOURCES = tst_knmi.c tst_utils.c
 tst_wrf_reads_SOURCES = tst_wrf_reads.c tst_utils.c
 tst_bm_rando_SOURCES = tst_bm_rando.c tst_utils.c
+tst_compress_SOURCES = tst_compress.c tst_utils.c
 
 # Removing tst_mem1 because it sometimes fails on very busy system.
 # Removing run_knmi_bm.sh because it fetches files from a server and
@@ -48,7 +50,7 @@ tst_bm_rando_SOURCES = tst_bm_rando.c tst_utils.c
 # in CI.
 TESTS = tst_ar4_3d tst_create_files tst_files3 tst_mem tst_wrf_reads	\
 tst_attsperf perftest.sh run_tst_chunks.sh run_bm_elena.sh		\
-tst_bm_rando
+tst_bm_rando tst_compress
 
 run_bm_elena.log: tst_create_files.log
 
diff --git a/nc_perf/run_gfs_test.sh.in b/nc_perf/run_gfs_test.sh.in
index 575ad37c9..8583f697c 100644
--- a/nc_perf/run_gfs_test.sh.in
+++ b/nc_perf/run_gfs_test.sh.in
@@ -7,6 +7,6 @@ if test "x$srcdir" = x ; then srcdir=`pwd`; fi
 . ../test_common.sh
 
 echo "*** Running tst_gfs_1..."
-@MPIEXEC@ -n 4 ./tst_gfs_data_1
+@MPIEXEC@ -n 4 ./tst_compress_par
 
 exit 0