diff --git a/debezium/perf/README.md b/debezium/perf/README.md
index 3b5e2d475ec..03ccaefd700 100644
--- a/debezium/perf/README.md
+++ b/debezium/perf/README.md
@@ -3,16 +3,16 @@
 The docker compose file in this directory is similar to
 the one in the ../demo directory, with additional
 images for running Materialize and Deephaven side-by-side
-with the same debezium and Kafka input stream being
+with the same Debezium and Kafka input stream being
 fed to both.
 
 Please see `../demo/README.md` for a general
 reference that also applies to this setup;
 in particular, the 'How to run' section should
-apply verbatim, substituing `debezium/demo`
+apply verbatim, substituting `debezium/demo`
 for this directory, `debezium/perf`.
 
-Please see the "Memmory and CPU requirements"
+Please see the "Memory and CPU requirements"
 section below; as this compose is performance
 analysis oriented, it has considerably
 larger requirements than our other
@@ -23,7 +23,7 @@ feature-oriented demos.
 On top of what is required for `../demo` (see
 `../demo/README.md`), the automated testing
 requires building the Deephaven Java client examples.
-At the toplevel directory of your git clone (`../..`), run:
+At the top level directory of your git clone (`../..`), run:
 
 ```
 ./gradlew java-client-session-examples:installDist
@@ -167,7 +167,7 @@ can make them execute their respective demo scripts.
 
 After running the commands above, both Materialize and
 Deephaven should be running semantically-equivalent
-demo scripts fed from debezium and kafka events
+demo scripts fed from Debezium and Kafka events
 (triggered by the loadgen script).
 
 You can increase or decrease the rate of events using
@@ -187,8 +187,7 @@ for instructions.
   SELECT
       total,
       to_timestamp(max_received_at) max_received_ts,
-      mz_logical_timestamp()/1000.0 AS logical_ts_ms,
-      mz_logical_timestamp()/1000.0 - max_received_at AS dt_ms
+      mz_logical_timestamp - 1000*max_received_at AS dt_ms
   FROM pageviews_summary;'  -U materialize -h localhost -p 6875
   ```
 
@@ -196,12 +195,12 @@ for instructions.
 
 The parameters used for images in the docker compose file in this
 directory are geared towards high message throughput.  While Deephaven
-itself is running with the same default configuration used for
-general demos (as of this writing, 4 cpus and 4 Gb of memory), the
-configurations for redpanda, mysql, and debezium are tweaked to reduce
-their impact in end-to-end latency and throughput measurements;
-we make extensive use of RAM disks (tmpfs) and increase some
-parameters to ones closer to production (e.g., redpanda's number
+itself is running with the same default configuration used for general
+demos (as of this writing, 4 CPU threads and an update cycle of 1
+second), the configurations for redpanda, MySQL, and Debezium are
+tweaked to reduce their impact in end-to-end latency and throughput
+measurements; we make extensive use of RAM disks (tmpfs) and increase
+some parameters to ones closer to production (e.g., redpanda's number
 of cpus and memory per core).  To get a full picture of the
 configuration used, consult the files:
 
@@ -210,13 +209,13 @@ configuration used, consult the files:
 * `docker-compose.yml`
 * `.env`
 
-Once started the compose will take around 6 Gb
-of memory from the host; as events arrive and
-specially if event rates are increased, it
-will increase to 10-16 Gb or more.
+Once started the compose will take around 3 GiB of memory from the
+host; as events arrive and specially if event rates are increased, it
+will increase.  To test rates of the order of 100k msg/sec, either of
+Deephaven or Materialize will need on the order of 12 GiB.
 
-For the mild initial rate (same as default demo in `../demo`),
-the compose will consume around 2 CPU threads
-(tested in a Xeon E5 2698 v4 CPU).
-For increased event rates (eg, 50,000 pageviews per second),
-CPU utilization will spike to 14 CPU threads or more.
+For the mild initial rate (same as default demo in `../demo`), once
+the demo code is loaded in both engines the compose will consume
+around 2 full CPUs (tested in a Xeon E5 2698 v4 CPU).  At high even
+rates, deephaven will take up to 3 CPU threads.  Materialize will take
+as many full cores as configured in their `-w` argument.
diff --git a/debezium/perf/run_experiment.sh b/debezium/perf/run_experiment.sh
index 0f846f0fdac..8d146ec8380 100755
--- a/debezium/perf/run_experiment.sh
+++ b/debezium/perf/run_experiment.sh
@@ -1,10 +1,18 @@
-#!/bin/sh
+#!/bin/bash
 
 set -eu
 
-if [ $# -ne 5 -o \( "$1" != 'dh' -a "$1" != 'mz' \) ]; then
+usage_and_exit() {
     echo "Usage: $0 dh|mz per_second_rate wait_seconds top_samples top_delay_seconds" 1>&2
     exit 1
+}
+
+if [ "$#" -ne 5 ]; then
+    usage_and_exit
+fi
+
+if [ "$1" != 'dh' -a "$1" != 'mz' ]; then
+    usage_and_exit
 fi
 
 engine="$1"
diff --git a/debezium/perf/sample_top.sh b/debezium/perf/sample_top.sh
index 671b6ce2efb..e3f0aed4d6c 100755
--- a/debezium/perf/sample_top.sh
+++ b/debezium/perf/sample_top.sh
@@ -2,8 +2,17 @@
 
 set -eu
 
-if [ $# -ne 3 -o \( "$1" != 'dh' -a "$1" != 'mz' \) ]; then
+usage_and_exit() {
     echo "Usage: $0 mz|dh rate nsamples delay_sec"
+    exit 1
+}
+
+if [ "$#" -ne 3 ]; then
+    usage_and_exit
+fi
+
+if [ "$1" != 'dh' -a "$1" != 'mz' ]; then
+    usage_and_exit
 fi
 
 engine=$1