diff --git a/.env.example b/.env.example
index 9eb87fc..3233082 100644
--- a/.env.example
+++ b/.env.example
@@ -32,6 +32,15 @@ LLAMA_GPU_LAYERS=999
 # to spare) or drop back to 131072 (128K), the last known-good value
 # (~25.6GB total, ~6GB headroom) — see docs/research/qwen3.8-27b-quant.md.
 LLAMA_CTX_SIZE=262144
+# Concurrent request slots. Was implicitly 4 (llama.cpp's compiled-in
+# default) with no flag set — under concurrent subagent fan-out, 4 requests
+# split the same GPU compute, so a large-context prefill can queue behind
+# others long enough to blow past OmniRoute's stream-idle timeout, which then
+# cancels the request (see issue-tracker notes on the timeout/cancel loop).
+# Dropped to 2 so each slot gets more compute and finishes prefill sooner;
+# raise back toward 4 if throughput (not latency) becomes the bottleneck
+# instead.
+LLAMA_PARALLEL=2
 
 # --- Lazytainer ---
 # Seconds of inactivity before llama-server is stopped. 900 = 15 min.
diff --git a/docker-compose.yml b/docker-compose.yml
index d353aab..59ce820 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -19,6 +19,7 @@ services:
       --port 8080
       --n-gpu-layers ${LLAMA_GPU_LAYERS:-999}
       --ctx-size ${LLAMA_CTX_SIZE:-131072}
+      --parallel ${LLAMA_PARALLEL:-2}
       --jinja
     # No published host port: llama-server is reached only via the omniroute
     # gateway on the ai-stack docker network now — see issue #15. Its