Luigi,
Are you talking about a client-only app (rich client or mobile or a web mashup), or your own server that will do the caching/manipulation of the content?
We don't really have a hard limit on the API usage (unless someone starts copying the entire wikipedia instead of using a dump file), especially if its a client-only approach with many clients accessing the data.
Assuming you are doing a client-only, your best bet is to probably use action=parse to get the HTML of the wiki text (it uses server-side cached IIRC). I am not sure about images - you can get image information through the api, but i don't recall if we have any server-side scaling exposed through it -- although it must be exposed somehow.
--Yuri