The basic problems are:

- "space scans", when Vanilla has to load every single snip (find) - as it is a CGI program, we can't keep any index/cached structures in memory.

- REBOL start-up overhead (for every single request, REBOL has to load, decompress mezzanines &c.).

- slow/massively shared hardware - dreamhost and friends are fine for static sites IMO, for dynamic applications a dedicated server (which can be pretty cheap nowadays) is of course the best solution.

Some options to improve performance:

- dedicated REBOL HTTP server: lots of work to make it stable enough for production use, usually impossible to set up on shared hosts à la dreamhost I think.

- dedicated REBOL server behind Apache (AJP13 &c.): hard to set up, usually impossible on shared hosts.

- "live" index structure generation: we do that in quite a few places (backlinks, recent-stores). hairy, introduces lots of dependencies.

- background index structure generation: quite a bit of work, requires cron access, usually impossible on shared hosts.

So the least work-intensive option is simply to get a dedicated server. A modern machine can easily handle dozens and dozens of Vanilla sites.