diff --git a/benchmarks/benchmarks/experimental/ugrid/regions_combine.py b/benchmarks/benchmarks/experimental/ugrid/regions_combine.py
index 8ebf210416..3b2d77a80a 100644
--- a/benchmarks/benchmarks/experimental/ugrid/regions_combine.py
+++ b/benchmarks/benchmarks/experimental/ugrid/regions_combine.py
@@ -11,8 +11,7 @@
   * minimal: enables detection of regressions in parts of the run-time that do
              NOT scale with data size.
   * large: large enough to exclusively detect regressions in parts of the
-           run-time that scale with data size. Aim for benchmark time ~20x
-           that of the minimal benchmark.
+           run-time that scale with data size.
 
 """
 import os
diff --git a/benchmarks/benchmarks/load/ugrid.py b/benchmarks/benchmarks/load/ugrid.py
index 8227a4c5a0..350a78e128 100644
--- a/benchmarks/benchmarks/load/ugrid.py
+++ b/benchmarks/benchmarks/load/ugrid.py
@@ -10,8 +10,7 @@
   * minimal: enables detection of regressions in parts of the run-time that do
              NOT scale with data size.
   * large: large enough to exclusively detect regressions in parts of the
-           run-time that scale with data size. Aim for benchmark time ~20x
-           that of the minimal benchmark.
+           run-time that scale with data size.
 
 """
 
@@ -39,7 +38,7 @@ def load_mesh(*args, **kwargs):
 
 
 class BasicLoading:
-    params = [1, int(4.1e6)]
+    params = [1, int(2e5)]
     param_names = ["number of faces"]
 
     def setup_common(self, **kwargs):
@@ -58,6 +57,9 @@ def time_load_mesh(self, *args):
 class BasicLoadingTime(BasicLoading):
     """Same as BasicLoading, but scaling over a time series - an unlimited dimension."""
 
+    # NOTE iris#4834 - careful how big the time dimension is (time dimension
+    #  is UNLIMITED).
+
     param_names = ["number of time steps"]
 
     def setup(self, *args):
@@ -75,7 +77,7 @@ class DataRealisation:
     warmup_time = 0.0
     timeout = 300.0
 
-    params = [1, int(4e6)]
+    params = [1, int(2e5)]
     param_names = ["number of faces"]
 
     def setup_common(self, **kwargs):
@@ -102,7 +104,7 @@ def setup(self, *args):
 
 
 class Callback:
-    params = [1, int(4.5e6)]
+    params = [1, int(2e5)]
     param_names = ["number of faces"]
 
     def setup_common(self, **kwargs):
diff --git a/benchmarks/benchmarks/save.py b/benchmarks/benchmarks/save.py
index bb1d2d8c82..3551c72528 100644
--- a/benchmarks/benchmarks/save.py
+++ b/benchmarks/benchmarks/save.py
@@ -10,8 +10,7 @@
   * minimal: enables detection of regressions in parts of the run-time that do
              NOT scale with data size.
   * large: large enough to exclusively detect regressions in parts of the
-           run-time that scale with data size. Aim for benchmark time ~20x
-           that of the minimal benchmark.
+           run-time that scale with data size.
 
 """
 from iris import save
diff --git a/docs/src/whatsnew/latest.rst b/docs/src/whatsnew/latest.rst
index 8dacecfd32..849d44a44c 100644
--- a/docs/src/whatsnew/latest.rst
+++ b/docs/src/whatsnew/latest.rst
@@ -232,10 +232,8 @@ This document explains the changes made to Iris for this release
    bin in the system PATH.
    (:pull:`4794`)
 
-#. `@trexfeathers`_ and `@pp-mo`_ fixed the CDL headers for
-   :mod:`iris.tests.stock.netcdf` to allow generation of NetCDF-4 files with an
-   unlimited time dimension.
-   (:pull:`4827`)
+#. `@trexfeathers`_ and `@pp-mo`_ improved generation of stock NetCDF files.
+   (:pull:`4827`, :pull:`4836`)
 
 #. `@rcomer`_ removed some now redundant testing methods. (:pull:`4838`)
 
diff --git a/lib/iris/tests/stock/file_headers/xios_2D_face_half_levels.cdl b/lib/iris/tests/stock/file_headers/xios_2D_face_half_levels.cdl
index ba3522b491..b135546f2d 100644
--- a/lib/iris/tests/stock/file_headers/xios_2D_face_half_levels.cdl
+++ b/lib/iris/tests/stock/file_headers/xios_2D_face_half_levels.cdl
@@ -55,8 +55,4 @@ variables:
 		:name = "${DATASET_NAME}" ;
 //		original name = "lfric_ngvat_2D_1t_face_half_levels_main_conv_rain"
 		:Conventions = "UGRID" ;
-
-// data
-data:
-  time_instant = 0 ;
 }
diff --git a/lib/iris/tests/stock/file_headers/xios_3D_face_full_levels.cdl b/lib/iris/tests/stock/file_headers/xios_3D_face_full_levels.cdl
index a87e3055c9..e4f32de7b7 100644
--- a/lib/iris/tests/stock/file_headers/xios_3D_face_full_levels.cdl
+++ b/lib/iris/tests/stock/file_headers/xios_3D_face_full_levels.cdl
@@ -58,8 +58,4 @@ variables:
 		:name = "${DATASET_NAME}" ;
 //		original name = "lfric_ngvat_3D_1t_full_level_face_grid_main_u3"
 		:Conventions = "UGRID" ;
-
-// data
-data:
-  time_instant = 0 ;
 }
diff --git a/lib/iris/tests/stock/file_headers/xios_3D_face_half_levels.cdl b/lib/iris/tests/stock/file_headers/xios_3D_face_half_levels.cdl
index f9c9c148dd..a193dbe451 100644
--- a/lib/iris/tests/stock/file_headers/xios_3D_face_half_levels.cdl
+++ b/lib/iris/tests/stock/file_headers/xios_3D_face_half_levels.cdl
@@ -58,8 +58,4 @@ variables:
 		:name = "${DATASET_NAME}" ;
 //		original name = "lfric_ngvat_3D_1t_half_level_face_grid_derived_theta_in_w3"
 		:Conventions = "UGRID" ;
-
-// data
-data:
-  time_instant = 0 ;
 }
diff --git a/lib/iris/tests/stock/netcdf.py b/lib/iris/tests/stock/netcdf.py
index c5ec5ce446..8a448f7d34 100644
--- a/lib/iris/tests/stock/netcdf.py
+++ b/lib/iris/tests/stock/netcdf.py
@@ -51,13 +51,13 @@ def ncgen_from_cdl(
             f_out.write(cdl_str)
     if cdl_path:
         # Create netcdf from stored CDL file.
-        call_args = [NCGEN_PATHSTR, cdl_path, "-k4", "-o", nc_path]
+        call_args = [NCGEN_PATHSTR, cdl_path, "-k3", "-o", nc_path]
         call_kwargs = {}
     else:
         # No CDL file : pipe 'cdl_str' directly into the ncgen program.
         if not cdl_str:
             raise ValueError("Must provide either 'cdl_str' or 'cdl_path'.")
-        call_args = [NCGEN_PATHSTR, "-k4", "-o", nc_path]
+        call_args = [NCGEN_PATHSTR, "-k3", "-o", nc_path]
         call_kwargs = dict(input=cdl_str, encoding="ascii")
 
     subprocess.run(call_args, check=True, **call_kwargs)