diff --git a/dev/release/verify-release-candidate.sh b/dev/release/verify-release-candidate.sh
index f4d2800eaf0..f1b11e3ec7f 100755
--- a/dev/release/verify-release-candidate.sh
+++ b/dev/release/verify-release-candidate.sh
@@ -519,7 +519,14 @@ test_and_install_cpp() {
     ${ARROW_CMAKE_OPTIONS:-} \
     ${ARROW_SOURCE_DIR}/cpp
   export CMAKE_BUILD_PARALLEL_LEVEL=${CMAKE_BUILD_PARALLEL_LEVEL:-${NPROC}}
-  cmake --build . --target install
+  # Expose conda's lib on the *fallback* path so the LLVM tools (llvm-link,
+  # llvm-ranlib) find libLLVM.*.dylib on flaky @rpath miss on macOS. (Fallback is searched last,
+  # so it doesn't shadow system libs like libiconv during the build or leak into tests.)
+  if [ "$(uname)" = "Darwin" ] && [ "${USE_CONDA}" -gt 0 ] && [ -n "${CONDA_PREFIX:-}" ]; then
+    DYLD_FALLBACK_LIBRARY_PATH="${CONDA_PREFIX}/lib:${DYLD_FALLBACK_LIBRARY_PATH:-/usr/local/lib:/usr/lib}" cmake --build . --target install
+  else
+    cmake --build . --target install
+  fi
 
   if [ ${TEST_CPP} -gt 0 ]; then
     LD_LIBRARY_PATH=$PWD/release:$LD_LIBRARY_PATH ctest \
@@ -779,10 +786,10 @@ test_source_distribution() {
 
   if [ "$(uname)" == "Darwin" ]; then
     NPROC=$(sysctl -n hw.ncpu)
-    export DYLD_LIBRARY_PATH=$ARROW_HOME/lib:${DYLD_LIBRARY_PATH:-}
+    export DYLD_LIBRARY_PATH=$ARROW_HOME/lib${DYLD_LIBRARY_PATH:+:${DYLD_LIBRARY_PATH}}
   else
     NPROC=$(nproc)
-    export LD_LIBRARY_PATH=$ARROW_HOME/lib:${LD_LIBRARY_PATH:-}
+    export LD_LIBRARY_PATH=$ARROW_HOME/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
   fi
 
   pushd $ARROW_SOURCE_DIR