diff --git a/bootstrap.sh b/bootstrap.sh
index 919c1a331448..6c5aa022db64 100755
--- a/bootstrap.sh
+++ b/bootstrap.sh
@@ -484,6 +484,14 @@ function bench {
 
 }
 
+function bench_if_enabled {
+  if [ "${BENCH_UPLOAD:-1}" == 0 ]; then
+    echo "Skipping benchmarks because BENCH_UPLOAD=0."
+    return
+  fi
+  bench
+}
+
 ### RELEASING ##########################################################################################################
 function versions {
   local noir_version anvil_version node_version cmake_version clang_version zig_version rustc_version wasi_sdk_version
@@ -750,14 +758,14 @@ case "$cmd" in
     export USE_TEST_CACHE=1
     export CI_FULL=1
     build_and_test full
-    bench
+    bench_if_enabled
     ;;
   "ci-full-no-test-cache")
     export CI=1
     export USE_TEST_CACHE=0
     export CI_FULL=1
     build_and_test full
-    bench
+    bench_if_enabled
     ;;
   "ci-chonk-input-update")
     export CI=1
diff --git a/ci.sh b/ci.sh
index 01d2e6aeb8b4..15123df7cccd 100755
--- a/ci.sh
+++ b/ci.sh
@@ -96,9 +96,17 @@ function multi_job_run {
   export AWS_SHUTDOWN_TIME_ARM=${AWS_SHUTDOWN_TIME_ARM:-90}
   export DENOISE=1
   export DENOISE_WIDTH=32
+  # Only the first full shard should run and upload the bench artifact. Heavy
+  # merge-queue modes already grind tests across many shards; repeating the
+  # post-test bench sweep on every shard is redundant and can overrun the queue.
+  local bench_primary=${1%% *}
+  export bench_primary
   run() {
     [ -n "${4:-}" ] && export REF_NAME=$4
-    PARENT_LOG_ID=$RUN_ID JOB_ID=$1 INSTANCE_POSTFIX=$1 ARCH=$2 exec denoise "bootstrap_ec2 './bootstrap.sh $3'"
+    local bench_upload=${BENCH_UPLOAD:-1}
+    [ "$1" != "$bench_primary" ] && bench_upload=0
+    PARENT_LOG_ID=$RUN_ID JOB_ID=$1 INSTANCE_POSTFIX=$1 ARCH=$2 BENCH_UPLOAD=$bench_upload \
+      exec denoise "bootstrap_ec2 './bootstrap.sh $3'"
   }
   export -f run