It actually works really well for render workloads like server-side rendering too, the reason for this is that rendering can be quite expensive and blocks the thread. With Serverless you're able to scale the renderer to infinite instances.

The cold-boot problem is an interesting one, it's heavily correlated to the serverless function size, this is why we implemented a complete output target for serverless, to output the smallest possible function that is completely standalone, no dependencies. This makes cold-boot considerably faster. For example on https://next-news.now.sh/ it's hard to tell the difference between cold-boot and warm functions.