Logged In: YES 
user_id=21627

The big win is in simplification of the code. Also, we are
not talking about 10k here. On Linux, sock_addr_t is 128
bytes. Take a look at set_error: it allocates 100 bytes for
an error buffer. Or sock_repr: 512 bytes for a buffer. Or
socket_gethostname: 1024 bytes. Or socket_gethostbyname_ex:
16384 bytes. socket_getaddrinfo: 30 bytes. os_init: 100 bytes

Or, looking at other modules: symtable.c:check_unoptimized:
300 bytes. Py_GetVersion: 250 bytes. PySys_SetArgv:
2*MAXPATHLEN+1 (on Linux, this is 8193 bytes). I could go on.

Don't worry about stack consumption. Or, if you do, analyse
the Python source code, and fix the big offenders first. 

Premature optimization is the root of all evil.