diff --git a/src/Lean/Elab/Frontend.lean b/src/Lean/Elab/Frontend.lean
index 0a248a42a7d4..132a7a32c3fb 100644
--- a/src/Lean/Elab/Frontend.lean
+++ b/src/Lean/Elab/Frontend.lean
@@ -145,7 +145,7 @@ def runFrontend
     : IO (Environment × Bool) := do
   let startTime := (← IO.monoNanosNow).toFloat / 1000000000
   let inputCtx := Parser.mkInputContext input fileName
-  let opts := Language.Lean.internal.cmdlineSnapshots.set opts true
+  let opts := Language.Lean.internal.cmdlineSnapshots.setIfNotSet opts true
   let ctx := { inputCtx with }
   let processor := Language.Lean.process
   let snap ← processor (fun _ => pure <| .ok { mainModuleName, opts, trustLevel }) none ctx
diff --git a/tests/bench/big_omega.lean b/tests/bench/big_omega.lean
new file mode 100644
index 000000000000..138db4cda456
--- /dev/null
+++ b/tests/bench/big_omega.lean
@@ -0,0 +1,40 @@
+/-!
+This benchmark exercises `omega` in a way that creates a big proof, exercising `instantiateMVars`
+and `shareCommonPreDefs` as well. In particular, running it with `internal.cmdlineSnapshots=false`,
+like the language server does, uncovered a significant slowdown in `instantiateMVars` (#5614).
+-/
+
+set_option maxHeartbeats 0
+theorem memcpy_extracted_2 (six0 s0x0 : BitVec 64)
+(h_six0_nonzero : six0 ≠ 0)
+(h_s0x1 : s0x1 + 16#64 * (s0x0 - six0) + 16#64 = s0x1 + 16#64 * (s0x0 - (six0 - 1#64)))
+(h_s0x2 : s0x2 + 16#64 * (s0x0 - six0) + 16#64 = s0x2 + 16#64 * (s0x0 - (six0 - 1#64)))
+(h_assert_1 : six0 ≤ s0x0)
+(h_assert_3 : six1 = s0x1 + 16#64 * (s0x0 - six0))
+(h_assert_4 : six2 = s0x2 + 16#64 * (s0x0 - six0))
+(n : Nat)
+(addr : BitVec 64)
+(h_le : (s0x0 - (six0 - 1#64)).toNat ≤ s0x0.toNat)
+(h_upper_bound : addr.toNat + n ≤ 2 ^ 64)
+(h_upper_bound₂ : s0x2.toNat + s0x0.toNat * 16 ≤ 2 ^ 64)
+(h_upper_bound₃ : s0x2.toNat + (16#64 * (s0x0 - (six0 - 1#64))).toNat ≤ 2 ^ 64)
+(h_width_lt : (16#64).toNat * (s0x0 - (six0 - 1#64)).toNat < 2 ^ 64)
+(hmemSeparate_omega : s0x1.toNat + s0x0.toNat * 16 ≤ 2 ^ 64 ∧
+  s0x2.toNat + s0x0.toNat * 16 ≤ 2 ^ 64 ∧
+    (s0x1.toNat + s0x0.toNat * 16 ≤ s0x2.toNat ∨ s0x1.toNat ≥ s0x2.toNat + s0x0.toNat * 16))
+(hmemLegal_omega : s0x1.toNat + s0x0.toNat * 16 ≤ 2 ^ 64)
+(hmemLegal_omega : s0x2.toNat + s0x0.toNat * 16 ≤ 2 ^ 64)
+(hmemSeparate_omega : s0x2.toNat + 16 % 2 ^ 64 * ((2 ^ 64 - (2 ^ 64 - 1 % 2 ^ 64 + six0.toNat) % 2 ^ 64 + s0x0.toNat) % 2 ^ 64) % 2 ^ 64 ≤
+    2 ^ 64 ∧
+  addr.toNat + n ≤ 2 ^ 64 ∧
+    (s0x2.toNat +
+          16 % 2 ^ 64 * ((2 ^ 64 - (2 ^ 64 - 1 % 2 ^ 64 + six0.toNat) % 2 ^ 64 + s0x0.toNat) % 2 ^ 64) % 2 ^ 64 ≤
+        addr.toNat ∨
+      s0x2.toNat ≥ addr.toNat + n))
+(hmemLegal_omega : s0x2.toNat + 16 % 2 ^ 64 * ((2 ^ 64 - (2 ^ 64 - 1 % 2 ^ 64 + six0.toNat) % 2 ^ 64 + s0x0.toNat) % 2 ^ 64) % 2 ^ 64 ≤
+  2 ^ 64)
+(hmemLegal_omega : addr.toNat + n ≤ 2 ^ 64) :
+ s0x2.toNat + (16#64 * (s0x0 - six0)).toNat ≤ 2 ^ 64 ∧
+  addr.toNat + n ≤ 2 ^ 64 ∧
+    (s0x2.toNat + (16#64 * (s0x0 - six0)).toNat ≤ addr.toNat ∨ s0x2.toNat ≥ addr.toNat + n) := by
+    bv_omega
diff --git a/tests/bench/speedcenter.exec.velcom.yaml b/tests/bench/speedcenter.exec.velcom.yaml
index 4282be76702a..b970bde71eae 100644
--- a/tests/bench/speedcenter.exec.velcom.yaml
+++ b/tests/bench/speedcenter.exec.velcom.yaml
@@ -368,3 +368,15 @@
   run_config:
     <<: *time
     cmd: lean big_do.lean
+- attributes:
+    description: big_omega.lean
+    tags: [fast]
+  run_config:
+    <<: *time
+    cmd: lean big_omega.lean
+- attributes:
+    description: big_omega.lean MT
+    tags: [fast]
+  run_config:
+    <<: *time
+    cmd: lean big_omega.lean -Dinternal.cmdlineSnapshots=false