To make myself clear I've done the calculation of the acumulative probability of a collision as the addresses are generated. This probability increases with the square of the number of generated addrsses. If we consider the most restrictive case: the 160 bit space of the public key hash, a 1% probabiliy of bitcoin addresses collision will be reached when 1.2x10^23 bitcoin addresses are generated. This means that a million machines generating 30 million addresses/second each would need a thousand years to reach a 1% probability of a single collision.
Of course even that overstates the scenario in the OP which is a "inuse key". Given the size of the key space in the highly improbable event of a collision it is very likely between two unfunded addresses created by the same mass creating entity.