No matter which way you slice it, you'll need a load of disk space to store those images. I wouldn't recommend using an index colour format because it'll look pretty darn awful.
The resolution is really dependent on what your clients will be doing with the imagery, and whether they'll be happy losing features smaller than 25m.
A way to reduce the size of the imagery is to use compression, either lossy (OK, technically 3-to-1 band conversion is lossy, but that's unacceptable lossyness, IMO) or lossless. Looking at the GDAL page on GeoTIFF, it supports LZW and DEFLATE lossless compression, as well as lossy JPEG compression. Client applications are possibly more likely to support native JPEG decompression (that's just educated guesswork on my part however) So I'd be inclined to divide up your imagery into tiles first, then apply JPEG compression to each tile.
As an alternative, as GeoServer maintains its own image metadata (which is stored directly in GeoTIFFs), you could skip the TIFF-JPEG part, and compress them directly as JPEG files using gdal_translate.
I'd also try the other compression types to see what sort of storage/speed trade-off you get and make your decision based on that.