It turns out that the interaction of the earth and the moon's gravity produces some non-intuitive results. The short answer appears to be that you can indeed hit the earth from the moon using less than the escape velocity of the moon.
The key, of course, is to consider that the escape velocity of the moon is calculated "in isolation", when in the case under consideration the earth's gravity, when acting in opposition to the moon's, will reduce the velocity requirements.
I've done a simple model of the system, notable mainly for the assumption that the moon's orbit is circular, and used a fairly coarse time step of 1 second and Verlet integration. In all cases, the cannon is assumed to be at the equator of the moon, firing perfectly retrograde. The frame of reference is the barycenter of the earth-moon system, and the motion of the earth is included.
EDIT - The following bold section turns out not to be true, but was an arifact of a bug in my simulation. Rather, there is a single window for a "direct" impact path, spanning 2512 to 2660 m/sec. Transit time for a perfect hit (path travels through earth center) is 418 ksec at 2574 m/sec launch. My apologies. I'm sure I've got it right this time. Positive.
In this case, there are two windows of launch velocity. The first, 2253 to 2260 m/sec, produces something like a direct fall to earth, with a trip duration of 424 to 440 ksec (about 6 days). The second window is larger, from 2470 to 2603 m/sec, and the resulting paths take a longer path, looping outwards and then back in. Trip times run from 418 to 424 ksec. Intermediate velocities (2260 to 2470) are similar to the successful paths, but they don't actually get within the radius of the earth, due in part to the earth's motion around the barycenter.
This model stops at 1 million seconds, so it's entirely possible that large values of launch velocity miss the earth the first time, then interact with the moon to produce a later impact, but I assume that this is not what is intended.
EDIT 2 - Since the consequence of falling from the moon's orbit and missing the earth is an orbit with apogee equal approximately to the moon's orbital radius, it's clear that there will be any number of possibilities for further interaction with the moon, and possible second tries. Depending on the details of the orbit (whether it keeps the same rotation as the moon's orbit or not), there is also the possibility of a slinghot resulting in a 2 km/sec projectile velocity, which will not be adequate to eject the projectile from the system, but will certainly cause a much larger orbit.
END EDIT 2
How can the projectile escape the moon's gravity when the velocity is less than the escape velocity? Simple. For velocities near escape velocity, the projectile spends a long time at large distances (compared to the moon's diameter), gradually slowing down, and at exactly escape velocity will take infinite time to reach zero radial velocity. During this long period of low velocity the earth's gravity will attract the projectile, and when the radial (lunar) velocity is on the order of the (terrestial) orbital velocity the result is acceleration towards the earth, which increases radial distance much more rapidly than would occur without the earth.