Grok starts up in just a hair less than 2 seconds on my hardware – it’s a MacBook Pro with 2.1 GHz Core 2 Duo, so perhaps a bit spiffier than average hardware. It is a lot more pleasant than the 15 or so seconds my Plone restarts take.

Stating all of your URL patterns in a central place a-la Routes has it’s appeal. It wouldn’t be too hard to write a custom Traverser for a Grok Application that resolved the rest of the URL in a Routes-like fashion. You would need to constrain your object heirarchy to just container/model, or require that all objects are indexed like Plone’s portal_catalog, or use SQLAlchemy. Pylons Controllers could map to Grok Containers, and Pylons Actions could map to Grok Views. It’s not exactly a clean fit though, so you’d probably want to change the semantics of your Routes-like API a fair bit. Or perhaps with a heavy dose of metaprogramming you could have map.connect() calls create Grok Traverser classes, although that would probably be way too funky …