Plone CMS: how heavy are search requests compared to typical CMS GET requests?
I fear that on a large site (0.5 milions of documents) enabling search capability is asking for DOS. If so, how this threat can be mitigated? Can search work on a different ZOE instance?
Plone's portal_catalog is rather efficient/fast/optimized. It's not like an SQL Query where you can construct searches that take minutes to complete.
The heavy part is usually "waking up" objects when presenting the search results, you should work as much as possible with the metadata (so called "brains") that the catalog returns. This is what Plone tries to do by default anyway.
But still, you can use a seperate ZEO instance for handling search request if you feel that this may be a bottleneck. Just make sure requests for /search and /search_form (or generically, /search*) end up at this specific ZEO instance. How you do this is rather specific to your current load balancing setup (apache, squid, nginx, etc)