I am also working on a project like this, though I haven't got my Arduino, yet. I'm just using a sound card for now... by merely touching the exposed wires plugged into my sound card with my fingers and intensive filtering in Audacity, I'm able to clearly detect the "R" spike of the QRS complex, and thus a heartbeat... but nothing else is discernible... I'm working on the amplifier.
It occurs to me that it is quite often that a hobbyist is touching things like exposed wires when messing with circuits... and nobody is very worried about it. But once we're actually measuring a heartbeat, it becomes dangerous?
I think there are many parts of this issue with safety and optocoupling/isolation and why it's perceived and/or actually as risky:
1) Very few accidents have occurred, but because they were caused in hospitals with plenty of people to blame (nurses, doctors, technicians, equipment manufacturers, etc) with lots of money behind them, it becomes a big legal liability problem. Hobbyists typically have no one to blame than themselves, and the rare accident that DOES happen may never be reported or the hobbyist, if they survive, may just brush it off or never even realize the danger they were in versus suing the hospital into the ground for an accident and causing a media sensation, even if the accident causes no (permanent) damage.
2) Early electrical equipment was overall far less safe than modern equipment... Electrocutions in general were far more common (for those with electricity) because of unsafe grounding practices that are now very rare except in antique equipment. Also, vacuum tubes were used for power supplies, amplifiers, etc. Vacuum tubes, even on amplifier circuits, almost always operate at high, dangerous voltages. Electrical wiring may not have been done by licensed professionals at the time, breakers and fuses were not as commonly placed as they are now, etc.... Thus, the need to be careful with ground was probably greater than it is today, although this sort of attitude can lead to complacency.
3) Hobbyists are merely ignorant of the danger that they are in. Also, they may use less capable power supplies, often even just batteries.
4) <conspiracy-theorist-hat>Medical device manufacturers want to keep the medical device market a very expensive market with little competition, even though the basic specifications of many medical devices could be replicated and exceeded by a hobbyist (and thus probably easily mass-produced) for very little money... Medical device manufacturers don't want medical devices to become commodities, which typically have very thin profit margins, so they lobby for stricter and more onerous safety standards, even if the safety isn't actually improved.</conspiracy-theorist-hat>
5) There are, in fact, electrodes attached directly to your skin by necessity, and the better the contact, the better the ECG signal, but also the greater the risk of lethal shock.
That being said, an opto/RF-isolated and battery-powered Arduino ECG would be pretty easy to do, just requires more time and complexity. But once you have that figured out and the design open-sourced, it should be pretty easy for others to replicate (can you do infrared serial with an arduino? it sounds doable, and would be a pretty cheap solution to opto-coupling). Also, you can get a pretty good measure of safety by merely hooking the Arduino into a laptop running on battery, not AC. Problem (mostly) solved!
Anyways, I've also though of adding a SPO2 sensor, a thermometer, and maybe some sort of digital stethoscope (so you could also do a sort of phonocardiogram... though it would take more bandwidth on the serial port, and thus would take more careful coding and more development time)... all in one device. Hook it up to an XBee or Bluetooth shield (or an SD card shield for recording), and you have yourself a very cheap device that can do all sorts diagnostic tests. Such a device, even with all those options, could probably be mass-produced for less than $100 (including bluetooth) and could integrate with cellphones (ubiquitous even among the poor in the third world) and the data shared all around the world with specialists.
I would really like to integrate an ultrasound device, but that is considerably beyond the processing power and/or bandwidth and/or sampling rate of an Arduino. BTW, what is the highest analog sampling rate of any hobby-like chip that is available that has a pretty accessible and high-bandwidth interface to a PC?