1) The primary reason for not implementing ESI would, in my humble opinion, be cost and complexity. Getting ESI right is actually really hard. You didn’t mention it – but you should consider it.

2) ESI done right, doesn’t impact performance significantly. I know – we’ve done benchmarks. It all depends on how well thought out the implementation is. ESI can be parsed at fetch-fime and a backend fetch is always a dead slow operation – so running through the document looking for ESI tags will not have a _measurable_ impact. Memory banwidth, which would be the limiting factor is usually over 10GByte/s today and unless you’re fetching gigabits you have lots to spare.

Besides, you will only ESI-process a small fraction of your content. Putting stuff together is just concatenating memory chunks and that is lightning fast.

3) Saying client side assembly is smarter isn’t always true. In web programming, relying on the client side to do something, to get stuff right, is dangerous. You might have noticed have many broken links to stuff like twitter are showing up these days.