org card: requests-per-GPU multiple to ~3.7x (nearly 4x), T=2 cached

Align the org profile throughput claim with the featured updated economics (was ~3x). Tokens (~49%), accuracy (+7.4pp), and storage (~5x) unchanged.

README.md

@@ -24,7 +24,7 @@ model to read the compressed representation. On ScienceQA the recipe holds or
 improves accuracy while shrinking the visual sequence:
 
 - **+8.4pp accuracy** over the un-spliced baseline at the accuracy tier (T=3, n=200)
-- At the compression tier (T=2, +7.4pp accuracy): **~49% fewer visual tokens** and **~3x more requests per GPU** in cached serving
+- At the compression tier (T=2, +7.4pp accuracy): **~49% fewer visual tokens** and **~3.7x (nearly 4x) more requests per GPU** in cached serving
 - **~5x smaller** stored visual representation
 
 The vision encoder and LLM weights are otherwise untouched; the compression sits