Well someone really lucky might end up with a freshly generated address that already has some coins on it. Right?!

What's even worse - investing isn't worth it either right now.
Value of bitcoin is growing quicker than just about all potential investments.
Since share price of your investment is based on the value (or expectation of future value) of the investment, it would have to grow faster than the bitcoin value which right now is close to impossible.
Otherwise you'll most likely be loosing money.

Well I understand that he sees the gains for F@H and has hopes to see the same with mining.
He just doesnt understand that there are fundamental differences.

As far as I can tell F@H and BOINC perform a rather complex series of calculations and therefore have quite complex code.
Also most of the code/algorithms is specific to F@H/BOINC, so more or less completely custom and not very well proven.
Both means there is a lot of potential for performance improvement, either in the code or in the algorithms themselves.

The calculation steps for bitcoin mining are relatively simple in terms of code complexity - they just need a lot of computing power to solve.
To my knowledge the majority of processing time is spent performing crypto algorithms over and over again.
Those are the same used in a lot of other applications for a long time already and are well proven and most potential efficiency gains have already been implemented.
So there is only little room for improvement - mostly reducing overhead from the communication with server/network and such things.

I wouldn't expect any improvements even close to what you saw for F@H with mining.

As far as ACICs are concerned - they are better at H/J and H/$, quite simply because they can't do anything else but mining with them. So you don't have to waste resources for those components that CPU/GPU have that are not necessary/not optimized for mining.