Vertex cache a bit more frequently.

Rely upon the minihashing a bit more.

GPU/GLES/TransformPipeline.cpp

@@ -672,9 +672,9 @@ void TransformDrawEngine::DoFlush() {
 							DecodeVerts();
 							goto rotateVBO;
 						}
-						if (vai->numVerts > 100) {
-							// exponential backoff up to 16 draws, then every 24
-							vai->drawsUntilNextFullHash = std::min(24, vai->numFrames);
+						if (vai->numVerts > 64) {
+							// exponential backoff up to 16 draws, then every 32
+							vai->drawsUntilNextFullHash = std::min(32, vai->numFrames);
 						} else {
 							// Lower numbers seem much more likely to change.
 							vai->drawsUntilNextFullHash = 0;