You don't need to scrape for each tweet. Scrape every active user every couple hours/days depending upon their tweet frequency. Not much point in having a database up to date every second since processing it in real time is almost impossible anyway.

Twitter would notice a major scraping operation, but if it's done correctly they wouldn't be able to distinguish between user IPs and bot IPs.

edit: Barracuda already did more than 10% of users just for a white paper: http://www.barracudanetworks.com/ns/news_and_events/index.ph...

150,000,000 registered users only takes 170 days at 10 users a second for a first pass. Focus on frequent tweeters for subsequent scrapes. Even among the ~20% of twitter accounts that are active, most don't need to be scraped daily, and the most active accounts are likely spam.