Steps to reproduce:

1. Open attached project
2. Change graphics API to 'Metal' for Mac Standalone
3. Build 'Elements Preview' scene
4. Attach the built app to Xcode project and observe the CPU/GPU frametime (this can only be done in macOS Sierra update, previous OSX versions don't have this functionality)

Repeat these steps with 5.5.0 (trunk) and metal/hlslcc/runtime branch, notice the big difference in both CPU and GPU frame time.

My results:
trunk:
CPU: 16 - 17ms
GPU: 16 - 18ms
metal/hlslcc/runtime:
CPU: 33 - 39ms
GPU: 37 - 38ms