diff --git a/scripts/Dockerfile.deps b/scripts/Dockerfile.deps
index a73993fb64..d5d89773fd 100644
--- a/scripts/Dockerfile.deps
+++ b/scripts/Dockerfile.deps
@@ -67,4 +67,14 @@ RUN ./build_linux.sh -u
 
 # Pinned dependencies in ./deps (OCCT 7.6 with ModelingAlgorithms enabled, OpenCV,
 # OpenVDB, Boost, wxWidgets, ...). This is the long step; it is baked into the image.
-RUN ./build_linux.sh -dr
+# -j 12, not the default all-cores: on 2026-08-21 an unbounded deps build plus an unbounded
+# app build put 42 GB of cc1plus on a 62 GB box and OOM-killed the host for 2h28m. 12 keeps
+# the compile fast while leaving the machine usable. Pair with `docker build --memory`.
+RUN ./build_linux.sh -dr -j 12
+
+# Compatibility symlink. This deps tree installs into deps/build/OrcaSlicer_dep/, while the
+# orcacad_buildcache volume was configured against the previous image, whose prefix was
+# deps/build/destdir/. Those paths are baked as absolute strings into the app's CMakeCache, so
+# without this symlink a rebuild on the new image reconfigures from scratch — hours of compile
+# to change one directory name. Same tree, two names.
+RUN ln -sfn /OrcaSlicer/deps/build/OrcaSlicer_dep /OrcaSlicer/deps/build/destdir
diff --git a/scripts/rig-build.sh b/scripts/rig-build.sh
index e749fff298..72e9fc3ace 100755
--- a/scripts/rig-build.sh
+++ b/scripts/rig-build.sh
@@ -42,13 +42,13 @@ fi
 # build produced a single object. The bounds, weakest to strongest:
 #   flock    — the lock path is shared by both forks on purpose, so they SERIALISE instead of
 #              summing. Peak is one build's worth no matter who else starts one.
-#   -j8      — ~8 TUs in flight, ~10-20 GB, and the box stays usable while it compiles.
-#              JOBS=n overrides for a machine with more headroom.
+#   -j12     — measured 1.17 GB average per cc1plus in the incident dump, so 12 in flight
+#              is ~14 GB typical and leaves the box usable. JOBS=n overrides.
 #   --memory — the actual guarantee. A runaway build hits its own cgroup limit and dies alone;
 #              the host never reaches global OOM again, whatever -j or flock do.
 #              --memory-swap equal to --memory forbids swap, which is what made ssh hang.
-JOBS="${JOBS:-8}"
-MEM="${MEM:-32g}"
+JOBS="${JOBS:-12}"
+MEM="${MEM:-40g}"
 LOCK=/tmp/orca-rig-build.lock
 
 exec 9>"$LOCK"