Roughly what size networks can you actually get useful information for? Every time I play with Gephi using any network of reasonable size (in my field), it's always a giant useless hairball.
50-100k nodes, as posters above said, it's very variable on the network structure, but for me I get highly information clusters. Try increasing your cutoff for interaction via edge weight might help.