docs: remove unmeasured performance claims and library comparisons

Remove all 2-5x/10-20x speed claims, Jython/GraalPy comparison table, and other unverified metrics from mkdocs documentation.

Author: howtis

docs/architecture.md

@@ -37,7 +37,7 @@ All communication happens via **MessagePack** binary protocol over stdin/stdout:
 | JSON | ~120 bytes | Medium | Built-in |
 | Pickle | ~40 bytes | Fast | Built-in (unsafe) |
 
-MessagePack is 2-5× faster than JSON for typical workloads and produces smaller payloads. Pickle is faster but not safe for cross-process communication.
+MessagePack provides a compact binary format with strong typing, making it well-suited for cross-process communication over stdin/stdout.
 
 ## Process Lifecycle
 

docs/examples/batch-operations.md

@@ -10,7 +10,6 @@ Demonstrates batch evaluation and execution — sending multiple requests in a s
 - `batchExec()` executes multiple statements in one round-trip
 - Pool batch operations return `CompletableFuture<List<PythonValue>>`
 - Timeout applies to the entire batch, not per-item
-- 10-20× faster than sequential `eval()` for many small operations
 
 ## Run It
 

docs/index.md

@@ -4,15 +4,6 @@ Run Python code from Java — with full CPython compatibility and zero setup.
 
 PythonEmbed embeds a real CPython interpreter in your JVM application via a subprocess + [MessagePack](https://msgpack.org/) binary protocol — just pure Java and CPython. All C extensions (numpy, scipy, torch) work out of the box.
 
-## Why PythonEmbed?
-
-| Concern | PythonEmbed | Jython/GraalPy |
-|---------|------------|----------------|
-| CPython compatibility | Full — numpy, scipy, torch, all C extensions | Limited to pure Python |
-| Crash isolation | Python segfaults never kill the JVM | Native crash kills the JVM |
-| Setup | One Gradle plugin line | Manual venv or JVM flag configuration |
-| Performance | MessagePack binary protocol, 2–5× faster than JSON | In-process (JNI-based) |
-
 ## Quick Look
 
 Add the Gradle plugin:

docs/usage/batch.md

@@ -44,7 +44,7 @@ Each `eval()`/`exec()` call involves:
 3. Read from stdout
 4. Deserialize response (MessagePack)
 
-Batch operations combine all requests into one MessagePack frame, eliminating N-1 round-trips. For many small evaluations, this can be **10-20× faster**.
+Batch operations combine all requests into one MessagePack frame, eliminating N-1 round-trips.
 
 ## Batch with Pool