Rendering performance of single large mesh vs multiple smaller meshes

I have seen questions similar to mine being asked but I did not find the answers to those particularly illuminating for my situation. So here goes...

I have a 3D maze lit by a directional light causing the maze to cast shadows on itself. Only a relatively small section of the maze is visible at any one time The maze mesh needs to be generated at runtime and I have tried two algorithms:

1. Generate a single mesh for the entire maze. The Game tab stats window shows a consistently high count of vertices being rendered i.e. there appears to be no frustum culling taking place. The frame rate seems okay for a moderately sized mesh but I'm concerned about performance for much larger mazes.

2. Split the maze mesh into multiple sections. The rendered vertex count is considerably lower and some dynamic batching takes place (only if the sections are sufficiently small). However there appears to be no reduction in render time.

So I am wondering what is the best strategy for optimal render performance. In theory I would like to create large mazes with a vertex count running into the hundreds of thousands so I really need to gain any performance advantage I can.

Note - I have previously written a longer post on this subject (Optimising rendering of a large maze mesh - batching issues etc). This gives much more detail on my project, the tests I've performed and the results I observed.