I have a 3.3 Volt PCI ethernet card working on a 5 Volt PCI slot. How is it possible?

Question

This is the design for PCI slots:

As you can see, there are 32/64 bits and 3.3/5 Volts.

For what I know:

• 32 bits PCI cards can work in 64 bits slots of the same voltage.
• 64 bits PCI cards can work in 32 bits slots of the same voltage.

I don't have an exact image, but this one I have found on internet is very close to the mine (mine has 4 RJ45, not 2) and shows how some of the PCI contacts can stay outside of the slot, and the card will work OK (maybe at a lower speed):

But I have found something strange in a computer (network server) inside an office: a 64 bits 3.3 Volts ethernet card running (working) on a 32 bits 5 Volts slot.

The card is a Quad Ethernet PCI-X 64 bits:

5740 IBM 03N5444 Quad Port 10/100/1000 Base-TX Ethernet PCI-X Ethernet Adapter

As shown in the PCI design image (see above), a 3.3 Volt PCI card should never fit in a 5 Volt PCI slot, as long as there is a notch (lock, latch or whatever we call it) so the card will never fit in the wrong slot.
But it seems that Attila the Hun was playing with this computer, because the card fitted in the slot. The trick has been to perform a simple cut, as shown in this image:

This is the photo of the surgery-affected card (sorry, my camera is not very good) :

Beside the funny thing of that trick, what is really awesome for me is that the card is working. And, according to employees in that office, it could be working perfect: the server is dedicated mostly to network traffic, a balance load of 3 DSL routers to 3 different LANs. Even when LAN traffic is very low on that office, there is a lot of internet traffic some days (maybe more than 100 users), so the 4-RJ45 network card should work in a rather intensive manner, or so I think.

I don't know for how long this server has been working with this Frankenstein network card (you know what happens in modern enterprises: everybody is a temporal worker :-P, so they can not tell me), but months at least, maybe more.

So, I have two questions:

• How is this possible? Why a 3.3 Volt PCI Card running on a 5 Volt PCI slot has not crashed?
• What could be expected to happen in a future? Strange network behaviors as long as Quad RJ45 circuitry gets melted? Damages to the full system (motherboard, RAM, CPU... etc)? Or... is it possible that nothing will happen?

The motherboard is an HP P5LP-LE (Leonite).

EDIT: I have access to this sort of "network overclocked" machine, and I think I can perform any test that would be needed, like intensive LAN traffic, replace LAN cards to compare, or obtaining info with any program. It runs Windows XP and Ubuntu Linux Server v14.04 LTS. During the nights (when there will be nobody in the offices) I could probably even install some more operating system.

EDIT-2: Here is the detail of the cut. It seems clear to me that it is handmade:

and the rear side:

Answer 1

The card itself is an 5740 IBM 03N5444 Quad Port 10/100/1000 Base-TX Ethernet PCI-X Ethernet Adapter which IBM states is a PCI-X 1.0a adapter that operates at 3.3 volts.

The motherboard itself is an HP P5LP-LE (Leonite) which has only PCI slots. More details from the manual itself here:

There are three 32-bit PCI slots on this motherboard. The slots support PCI cards such as a LAN card, SCSI card, USB card, and other cards that comply with PCI specifications.

With that in mind, this answer on Super User addresses the issue of using PCI-X cards in plain PCI slots:

Yes, as long as the PCI slot is a 2.x or later, PCI 1.0 was 5v while PCI 2.x was 3.3v - which is electrically compatible with PCI-X.

So knowing that, it does seem that someone manually “hacked” and extra notch to allow the 5740 IBM 03N5444 Quad Port 10/100/1000 Base-TX Ethernet PCI-X Ethernet Adapter to be keyed as a “universal” PCI card.

What are the ramifications of this? Honestly, unclear. This site does give some insight; the 3.3 Volt designation and 5 Volt designation refers to the voltage level of clock & timing signals and not power supply values:

The PCI specification defines two basic types of expansion connectors that may be found on a motherboard — one for systems with 5-Volt signalling levels, and the other for systems using 3.3-Volt signalling levels. This specifies the voltage level of the various clock and timing signals, but not necessarily the power supply voltage. A particular card may require both 5-Volt and 3.3-Volt power supplies irrespective of its signaling-level voltage.

Also more details from this post on what 3.3 Volt versus 5 Volt actually means; emphasis is mine:

33 MHz cards run on 5 volt signalling. 66 MHz cards use 3.3 volt signalling. (They both are powered by five volts; only the signalling on the bus is different.) The first "notch" (keyway) in a 66 MHz card is in a different place than that in a 33 MHz card; this is the different "shape" alluded to above. One would think that this would prevent you from putting a 33 MHz, five-volt only card in a 66 MHz, 3.3 volt slot. Or vice versa.

So that basically means that electrically, the card should be stable. But clock & timing issues? That might be the rub. You say the card is working, but perhaps it’s being inadvertently over clocked? Thus causing issues? Or maybe it’s working because the HP P5LP-LE (Leonite) motherboard is properly throttling down the voltage on that slot to 3.3 Volts? My gut feeling is it would be the later. But still have not found any details in the PCI specifics of the slots on that motherboard other than the 32-bit quote above.

My advice? If you feel a bit unsure about this setup, find a plain PCI Ethernet card with 4 ports for a replacement. Should cost less than $50 or even less than that and will be more stable in the long run.