@anx said "Test it" (in bold, nonetheless) and so I did.
Well, not really, since I don't have the time for that. Since I suspected that the different solution may be apart by orders of magnitude, I saved myself the pain of: Actually configuring things and setting up another system to run a load-test. Especially since the specs of my reverse proxy hint that the backend servers are not high volume.
Instead I installed the packages (just apt install <App>) and did free -m && service <APP> start && free -m && service <APP> stop && free -m. Waiting a few seconds between the steps. I also used Debian 11, which, at the time of writing, is still the testing branch.
Very rudimentary, I know, but it served my purpose. This is the result, formatted as: Application (Disk space used, Additional packages installed by apt, memory used)
sniproxy (298 kB, 2, 1-2 MB)
apache2 (8,634 kB, 6, 3-4 MB)
nginx (20,500 kB, 20, 4-6 MB)
haproxy (4,315 kB, 1, 34-35 MB)
I've included apache since I assume that it also includes a similar functionality, even though it was not mentioned in the linked post. Be aware, this is not scientifically solid, given that I just used free and memory usage by the operating system can of course contribute to the changes.
So, in the UNIX tradition, use a tool, which does one job only (and hopefully does it good). So if you just need a lightweight Layer 4 proxy with SNI support, go with sniproxy. The linked post also mentioned, that it's easy to configure.
