Timed by my own fair hand...
Turbo C 2.01 (1988), 5s in the IDE, 7s from the command line
Borland C 2.0 (1991), 15s, or 13s using precompiled headers, both in the IDE, I didn't have the command line version on the CF card.
This was run from compact flash, I am pretty sure it runs faster from RAM drive but there's not enough room on this one, my other one, when I find it, not only had 6MB of RAM but it also had a double-speed upgrade. Woooo!
Back in the day, I used to write my C in the Borland IDEs, compile it up for syntax errors, then build it "properly" in the company standard, Microsoft C 6, because the Borland tools were so much faster. IF you ever used Microsoft's Programmer's Workbench (PWB), you'll know what I mean. It took Microsoft several years to catch up. They introduced QuickC but it was so crippled compared to Borland's offerings. The straw that broke the camel's back in the C wars on Microsoft OS's was Windows, where they'd had to introduce so many kludges to make the environment work that the competitors' compilers and IDEs needed to be redesigned and rebuilt pretty much from scratch. Even then to debug early Windows stuff you had two choices, either Symdeb or Codeview, with a second mono monitor (not possible on the same screen) or with symdeb you could do it over a serial connection. I wrote a special Windows VGA display driver to allow Codeview and Windows to run concurrently on a split screen, that saved a lot of grief, it was pretty unusual to have two monitors in those days. Memories...