Scalability vs. Performance: it isn't a battle

However, the "sweet spot" for acquisitions seems to be when your business reaches the 20-100 server range. LiveJournal, Wikipedia, Meebo, Flickr, etc. all seem to be around that size.

That's precisely the range where scalability does matter and performance doesn't. You've outgrown the put-web-and-DB-on-separate-boxes-and-replicate approach, so you actually have to think about partitioning & shared-nothing. But 30 servers @ $200/month/server is only about $6000/month, the cost of a single employee. At this stage, it doesn't really make sense to sacrifice developer hours for operational costs.

If you get to be really big, it makes a whole lot more sense. 100,000 servers @ $200/month/server = $20M/month, more than the purchase price of a small smartup. At this point, it makes a lot of sense to put a team of a half-dozen or so programmers onto the problem of speeding up individual server performance. If you can double performance, you save $10M/month, which'll pay wages for close to 1000 programmers.

There's a reason why Google writes all their performance-critical stuff in C++.