It's even more interesting when you consider that it ran under MS-DOS, which had no dynamic linking --- it's a statically linked binary.

The touch command under OS X Lion (44,016 bytes).

...and I bet that one is dynamically linked.

I believe the reason why it's so fast is because it has no optimiser, and generates code as it parses. Pascal is also one of the easier languages to compile.

What is all the weight in that "touch" binary? It can't be code. My guess would be that the format is just not optimized for size and contains a bunch of boilerplate records whose minimum size is in the kilobytes.

    $ bloaty `which touch` -d segments,sections --domain=file
       FILE SIZE
    ---------------
     55.1%  19.6Ki    __LINKEDIT
       91.7%  18.0Ki    Code Signature
        2.7%     544    Symbol Table
        1.9%     376    Lazy Binding Info
        1.6%     328    String Table
        1.2%     232    Indirect Symbol Table
        0.6%     112    Binding Info
        0.2%      32    Export Info
        0.1%      24    Function Start Addresses
        0.0%       8    Rebase Info
        0.0%       8    Table of Non-instructions
        0.0%       0    [__LINKEDIT]
     22.2%  7.90Ki    __TEXT
       35.0%  2.77Ki    __TEXT,__text
       33.6%  2.65Ki    [__TEXT]
       17.2%  1.36Ki    [Mach-O Headers]
        4.2%     339    __TEXT,__cstring
        3.4%     274    __TEXT,__const
        3.3%     266    __TEXT,__stub_helper
        1.9%     150    __TEXT,__stubs
        1.5%     120    __TEXT,__unwind_info
     11.2%  4.00Ki    __DATA
       94.9%  3.80Ki    [__DATA]
        4.9%     200    __DATA,__la_symbol_ptr
        0.2%       8    __DATA,__data
     11.2%  4.00Ki    __DATA_CONST
       98.4%  3.94Ki    [__DATA_CONST]
        1.6%      64    __DATA_CONST,__got
      0.3%     104    [Mach-O Headers]
    100.0%  35.6Ki    TOTAL

Those [__TEXT], [__DATA], and [__DATA_CONST] sections are the part that is lost to padding, so 10.4Ki or so.

Disclosure: I am the author of Bloaty.