Increasing the cache limit is easy to do in the UI:

This setting is stored in the user’s preference file (~/.B.blend on Linux systems) and, therefore, applies to all projects. However, rendering from the command line requires using the -b argument, which explicitly ignores the user’s preference file. Apart from editing and re-compiling the Blender source code, I found only one way to increase the memory cache limit.

The trick is to use a Python script, like below, to change the user preferences before rendering the scene:

import sys
import bpy

def main():
	bpy.config.sequenceMemCacheLimit=4096

if __name__ == '__main__':
	main()

Add the script to the command line like this:

./bin/blender -b sample-static-text.blend -P render_settings.py -a

Note that the order of arguments matters because first we need to load the scene (with -b), then change settings through Pythons (using the -P option) and finally render the animation (with -a). Also, you need at least version 2.49a of Blender for this to work.

For reference, here’s is the the source file which hardcodes the 32MB limit:

intern/memutil/intern/MEM_CacheLimiterC-Api.cpp:

...
static intptr_t & get_max()
{
        static intptr_t m = 32*1024*1024;
        return m;
}
...