So lets do the math. vanitygen can generate about 1 MKey/s with a few tweaks (e.g. try positive/negative, compressed and uncompressed keys at the same time) on today's computer. Lets assume that using ASIC technology (which you have to build from scratch, since SHA-256 is not enough to generate private/public key pairs) you can get a speedup of a trillion (the current Bitcoin network has significantly less than a trillion times more computing power than Satoshi's computer). Lets assume there are about 100 million addresses that currently have a non-zero balance (I think it's less). There are only 2^160 addresses (since we hash to 160 bits). So you need on average
2^160 addresses / (100 million non-empty addreses) / a trillion speedup / 1 million (keys/s) /31.5 million (s/year) = 464*10^12 years/key
This are around 465 trillion years to find a single key for a non-empty address, if I'm not mistaken.
Okay, if Moore's law will still hold for 75 years, then this method might become feasible -- if you want to invest as much money as all miners together and let the computer work for half a year to find a fraction of a Bitcoin in some random address.
BTW, there are faster ways than brute-forcing all addresses. You could take an address with lot of funds where the public key is known (e.g. Bitstamp's cold wallet) and use the big-step, baby-step algorithm that "only" has complexity 2^128. That may be 100 times faster.
And since you asked for quantum computers: The current quantum computers can probably not be used, but if you have a real 65000-qbit quantum computer that can do complex computations without decoherence for a few seconds, then you can probably crack a public key in a few seconds.