The sort of scenario that would convince me that things are as you suggested in your first post:

1) Bidder A places a bid which puts him in the "high bidder" position at less than his maximum bid

2) Bidder B places a bid late in the auction, and for less than Bidder A's maximum bid

3) Bidder B wins the auction, and at a closing price that is lower then Bidder A's maximum bid.

I have never seen evidence of this. If you can ever find a place where such is reliably documented, then please let me know. If this were truly the case, it would seem to be a terribly important issue to eBay; any bug that keeps the closing price artificially low would be worth fixing.

Also, please consider that every bid is part of the proxy bidding system, including your "snipe bid". When your bid is placed, the system does not simply set the current high bid to the amount you bid. It has to compare your maximum bid to the maximum bid already placed by the current high bidder, see which one is higher, and then set the new high bid to the lower of the two plus the minimum bid increment (or to the maximum bid, if it's not high enough to cover the minimum bid increment).

I know it may be "hard to believe" that someone bid very close to your max bid amount, but it happens. I remember the feeling that something funny was going on once when I won an auction for exactly my maximum bid, and I had bid an oddball amount (like $73.77, for example). Looking at the bid history, it turned out that the next highest bid was less than mine by less than the minimum bid increment, and the proxy system had simply bid me up to my max. Seemed like voodoo, but it wasn't, just close bidding.

I also saw an auction a while ago (was watching it to decide how to price something similar that I was planning on selling) in which two bidders bid exactly the same amount ($800) just before close. Great news for the seller, as the next higest bid before that had been several hundred dollars less. Tough news for the guy who bid just a smidge later, as the earlier bid wins if they're exactly the same. Should have used that $0.77...