I found the node of top witness was running c5.large of AWS which has 4g ram

That's pretty strange. Maybe I can try and get in touch with him/her to understand this better. I went in for a trial and error method to find the bare minimum to run my node. So, I jumped from 4G to 8G to 16G ram. Finally, saw one of the witnesses running on a 30G ram so settled that for my backup node.

As you might know I am running the witness fir blurt in my rpi4 8g

I haven't personally checked out their code. But from a quick search, it seems like they have a much lower requirement than Steem. Need to check their post FORK changes to see what let them reach that point.

PS: The CPU usage by the container on my main node hits 65-70% regularly. So, there is definitely a lot of load on the CPU.