Since mining works fine in a 1x PCIe slot, why not go even slower with the conventional PCI bus?

I bought one of these adapters for testing ($28 incl. shipping on Ebay) and it seems to work fine. HD5870 at 900 MHz gives 354 Mhash/s, which is no different from running it in a 16x PCIe slot. Now I can fit even more cards in this single mobo

There are just a few practical considerations. PCI is a shared bus, so performance may go down with more than one of these adapters. Also, PCI only gives 25 W of power, vs. 75 W of PCIe, but this does not seem to be a problem when the card is powered externally. Mounting the card is yet another issue, since this is basically a riser card.

Also, X refused to start in the usual way, but it worked after setting BusID explicitly. In my case, it was

BusID  "PCI:6:0:0"

The exact number can be seen from lspci, in a slightly different format like 06:00.0 in this case. The line goes in the Device section of xorg.conf.

(using Linux 2.6.37.5, fglrx 10.12, SDK 2.1)

I propose an idea for cheating in a mining pool. Please comment on the viability and suggest countermeasures if needed.

The current pool by slush uses standard getwork miners with an artificially low difficulty. Such succesful hashes are used to gauge the level of contribution from a given miner. Now suppose that whenever a miner finds a block that satisfies the actual Bitcoin difficulty, they return it to their own local bitcoind instead of the pool. The pool continues to get measures of contribution, but the actual contribution in terms of 50 BTC blocks goes to the miner. In addition, they get their "share" of the pool based on their "contribution".

Of course, both the ethical and practical problem is that the pool will not generate anything if everyone does this. But suppose that some are already doing it. In the long run, statistics may show who is cheating, but it is not obvious in the short term.