Here's my Lemmy docker stats for lemmy.mods4ever.com

CONTAINER ID   NAME               CPU %     MEM USAGE / LIMIT     MEM %     NET I/O           BLOCK I/O         PIDS 
84102db7f1d2   lemmy_proxy_1      0.00%     9.27MiB / 7.566GiB    0.12%     6.68GB / 2.9GB    188kB / 463kB     5 
4e6a8c728e47   lemmy_lemmy-ui_1   0.00%     335.3MiB / 7.566GiB   4.33%     1.94GB / 5.46GB   10.1MB / 0B       11 
dc58b8518b9d   lemmy_lemmy_1      0.06%     265.4MiB / 7.566GiB   3.43%     16GB / 6.13GB     5.46MB / 0B       10 
0e08ceb5f404   lemmy_pictrs_1     0.14%     52.73MiB / 690MiB     7.64%     12.1MB / 555MB    48.5MB / 2.88GB   20 
c275f58231d4   lemmy_postgres_1   0.00%     574.9MiB / 1000MiB    57.49%    3.23GB / 14.8GB   30.2MB / 4.46GB   37 
1f3ac56b2a06   lemmy_postfix_1    0.00%     5.168MiB / 7.566GiB   0.07%     173kB / 126B      926kB / 77.8kB    7 

I can't find/remember the post but recently an admin shared some resource usage graphs. The thing that jumped out at me was their Lemmy instance was using 32 GB of ram while their PieFed instance was using 2 GB of ram. Maybe a little more, but it was low single digits.

Maybe their piefed graph did not include postgresql, which should be 2 GB just by itself... That would bring the total up to around 4 or 5 GB. piefed.social uses between 4 and 5 GB, too.

CPU usage was quite similar between the two.

I run a personal lemmy instance and two personal piefed instances, so I was just doing some comparisons. My instances are mainly used for development and testing, so they are only subscribed to a handful of communities and just have one active user.

You are correct that when it comes to performance, like snappiness and responsiveness, the database is probably going to be the bottleneck. Unless you are scaling up to a huge degree, I would be surprised to see meaningful differences in the number of requests that could be handled due to language differences between rust and python. Yes, python is an interpreted language, but most of the libraries you are using are basically calling other system libraries written in a language like C, and the program can execute way faster than your database I/O can give it data to process anyway.

Here is my usage summary. The lemmy instance has been running for about 1.5 years while the piefed instance has been running for just shy of a year now. I have only included the memory usage and disk since I don't think either is really CPU hungry or bound in my use case.