Yes. That would require a massive reprogramming of the guidance computer by the person taking over the silo. Most ICBMs are programmed to go up, tilt towards their target, and then slo-o-o-owly pitch over to the desired flight path angle. They also have a very very simple algorithm – sometimes hard coded – to run through the first minute of flight. Many of them can’t effectively manage a loop because their gyros would lock up. Also, tilting too quickly while still inside the atmosphere (~100km) puts a bending force on the airframe, which is designed to survive mostly compression-only loads. Turn too sharply at too low of an altitude and you’ll snap the missile like a matchstick using only aerodynamic loads – a neat trick, but what a waste!
It would be possible to design a two-or-three-stage ICBM guidance program to drive the final stage backwards and let gravity take over, but this would require weeks or months of simulations to get right, a deep understanding of the guidance hardware, and the ability to completely reprogram the guidance software. For obvious reasons, most ICBMs don’t make this easy; some older Soviet ICBMs make this flat-out impossible. Assuming you could do this, you’d need to drop the reentry vehicle on a shallow enough path towards the target to avoid burning itself up. So you’ve got to go far enough downrange that the RV’s return path is shallow, but close enough that the second and/or third stages working together can get the RV back to the origin.
Assuming no modification to the guidance system, it basically depends what the guidance algorithm is, how good the RV’s heat shields are, and how much velocity the first stage imparts. A very modern ICBM with a fast-burning first stage could perform an extremely steep first-stage burn, and as long as the reentry vehicle could handle the steep reentry (not a given!) then its minimum range could be very short indeed.
As for Chris Booth’s question, an ICBM (according to the U.S. Missile Defense Agency’s definition) can deliver its payload to a range of at least 5,500km. It does not have to exceed that range on every flight, nor carry the same payload on every flight – the systems are classed according to their potential. You could take a system that could only go 5,000km (an IRBM) and replace the payload with a bowling ball, and you’d probably have an ICBM.
Oh, and one last anecdote. You can actually get an ICBM to launch and still hit its own silo if you use the “cold launch” method, but it’s not a true launch. Anecdote from Astronautix.ru:
When they say “the rest of the launch sequence failed,” they mean that the ICBM was lifted several meters clear of the silo by the “mortar”, and then its guidance computer told the engines to start. They didn’t. The missile achieved a very very low apogee and then returned to its silo tail-first.